xref: /openbmc/linux/arch/x86/kernel/cpu/bugs.c (revision 7f8256ae)
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 #include <asm/cpu.h>
37 
38 #include "cpu.h"
39 
40 static void __init spectre_v1_select_mitigation(void);
41 static void __init spectre_v2_select_mitigation(void);
42 static void __init retbleed_select_mitigation(void);
43 static void __init spectre_v2_user_select_mitigation(void);
44 static void __init ssb_select_mitigation(void);
45 static void __init l1tf_select_mitigation(void);
46 static void __init mds_select_mitigation(void);
47 static void __init md_clear_update_mitigation(void);
48 static void __init md_clear_select_mitigation(void);
49 static void __init taa_select_mitigation(void);
50 static void __init mmio_select_mitigation(void);
51 static void __init srbds_select_mitigation(void);
52 static void __init l1d_flush_select_mitigation(void);
53 
54 /* The base value of the SPEC_CTRL MSR without task-specific bits set */
55 u64 x86_spec_ctrl_base;
56 EXPORT_SYMBOL_GPL(x86_spec_ctrl_base);
57 
58 /* The current value of the SPEC_CTRL MSR with task-specific bits set */
59 DEFINE_PER_CPU(u64, x86_spec_ctrl_current);
60 EXPORT_SYMBOL_GPL(x86_spec_ctrl_current);
61 
62 static DEFINE_MUTEX(spec_ctrl_mutex);
63 
64 /* Update SPEC_CTRL MSR and its cached copy unconditionally */
65 static void update_spec_ctrl(u64 val)
66 {
67 	this_cpu_write(x86_spec_ctrl_current, val);
68 	wrmsrl(MSR_IA32_SPEC_CTRL, val);
69 }
70 
71 /*
72  * Keep track of the SPEC_CTRL MSR value for the current task, which may differ
73  * from x86_spec_ctrl_base due to STIBP/SSB in __speculation_ctrl_update().
74  */
75 void update_spec_ctrl_cond(u64 val)
76 {
77 	if (this_cpu_read(x86_spec_ctrl_current) == val)
78 		return;
79 
80 	this_cpu_write(x86_spec_ctrl_current, val);
81 
82 	/*
83 	 * When KERNEL_IBRS this MSR is written on return-to-user, unless
84 	 * forced the update can be delayed until that time.
85 	 */
86 	if (!cpu_feature_enabled(X86_FEATURE_KERNEL_IBRS))
87 		wrmsrl(MSR_IA32_SPEC_CTRL, val);
88 }
89 
90 noinstr u64 spec_ctrl_current(void)
91 {
92 	return this_cpu_read(x86_spec_ctrl_current);
93 }
94 EXPORT_SYMBOL_GPL(spec_ctrl_current);
95 
96 /*
97  * AMD specific MSR info for Speculative Store Bypass control.
98  * x86_amd_ls_cfg_ssbd_mask is initialized in identify_boot_cpu().
99  */
100 u64 __ro_after_init x86_amd_ls_cfg_base;
101 u64 __ro_after_init x86_amd_ls_cfg_ssbd_mask;
102 
103 /* Control conditional STIBP in switch_to() */
104 DEFINE_STATIC_KEY_FALSE(switch_to_cond_stibp);
105 /* Control conditional IBPB in switch_mm() */
106 DEFINE_STATIC_KEY_FALSE(switch_mm_cond_ibpb);
107 /* Control unconditional IBPB in switch_mm() */
108 DEFINE_STATIC_KEY_FALSE(switch_mm_always_ibpb);
109 
110 /* Control MDS CPU buffer clear before returning to user space */
111 DEFINE_STATIC_KEY_FALSE(mds_user_clear);
112 EXPORT_SYMBOL_GPL(mds_user_clear);
113 /* Control MDS CPU buffer clear before idling (halt, mwait) */
114 DEFINE_STATIC_KEY_FALSE(mds_idle_clear);
115 EXPORT_SYMBOL_GPL(mds_idle_clear);
116 
117 /*
118  * Controls whether l1d flush based mitigations are enabled,
119  * based on hw features and admin setting via boot parameter
120  * defaults to false
121  */
122 DEFINE_STATIC_KEY_FALSE(switch_mm_cond_l1d_flush);
123 
124 /* Controls CPU Fill buffer clear before KVM guest MMIO accesses */
125 DEFINE_STATIC_KEY_FALSE(mmio_stale_data_clear);
126 EXPORT_SYMBOL_GPL(mmio_stale_data_clear);
127 
128 void __init check_bugs(void)
129 {
130 	identify_boot_cpu();
131 
132 	/*
133 	 * identify_boot_cpu() initialized SMT support information, let the
134 	 * core code know.
135 	 */
136 	cpu_smt_check_topology();
137 
138 	if (!IS_ENABLED(CONFIG_SMP)) {
139 		pr_info("CPU: ");
140 		print_cpu_info(&boot_cpu_data);
141 	}
142 
143 	/*
144 	 * Read the SPEC_CTRL MSR to account for reserved bits which may
145 	 * have unknown values. AMD64_LS_CFG MSR is cached in the early AMD
146 	 * init code as it is not enumerated and depends on the family.
147 	 */
148 	if (cpu_feature_enabled(X86_FEATURE_MSR_SPEC_CTRL)) {
149 		rdmsrl(MSR_IA32_SPEC_CTRL, x86_spec_ctrl_base);
150 
151 		/*
152 		 * Previously running kernel (kexec), may have some controls
153 		 * turned ON. Clear them and let the mitigations setup below
154 		 * rediscover them based on configuration.
155 		 */
156 		x86_spec_ctrl_base &= ~SPEC_CTRL_MITIGATIONS_MASK;
157 	}
158 
159 	/* Select the proper CPU mitigations before patching alternatives: */
160 	spectre_v1_select_mitigation();
161 	spectre_v2_select_mitigation();
162 	/*
163 	 * retbleed_select_mitigation() relies on the state set by
164 	 * spectre_v2_select_mitigation(); specifically it wants to know about
165 	 * spectre_v2=ibrs.
166 	 */
167 	retbleed_select_mitigation();
168 	/*
169 	 * spectre_v2_user_select_mitigation() relies on the state set by
170 	 * retbleed_select_mitigation(); specifically the STIBP selection is
171 	 * forced for UNRET or IBPB.
172 	 */
173 	spectre_v2_user_select_mitigation();
174 	ssb_select_mitigation();
175 	l1tf_select_mitigation();
176 	md_clear_select_mitigation();
177 	srbds_select_mitigation();
178 	l1d_flush_select_mitigation();
179 
180 	arch_smt_update();
181 
182 #ifdef CONFIG_X86_32
183 	/*
184 	 * Check whether we are able to run this kernel safely on SMP.
185 	 *
186 	 * - i386 is no longer supported.
187 	 * - In order to run on anything without a TSC, we need to be
188 	 *   compiled for a i486.
189 	 */
190 	if (boot_cpu_data.x86 < 4)
191 		panic("Kernel requires i486+ for 'invlpg' and other features");
192 
193 	init_utsname()->machine[1] =
194 		'0' + (boot_cpu_data.x86 > 6 ? 6 : boot_cpu_data.x86);
195 	alternative_instructions();
196 
197 	fpu__init_check_bugs();
198 #else /* CONFIG_X86_64 */
199 	alternative_instructions();
200 
201 	/*
202 	 * Make sure the first 2MB area is not mapped by huge pages
203 	 * There are typically fixed size MTRRs in there and overlapping
204 	 * MTRRs into large pages causes slow downs.
205 	 *
206 	 * Right now we don't do that with gbpages because there seems
207 	 * very little benefit for that case.
208 	 */
209 	if (!direct_gbpages)
210 		set_memory_4k((unsigned long)__va(0), 1);
211 #endif
212 }
213 
214 /*
215  * NOTE: This function is *only* called for SVM, since Intel uses
216  * MSR_IA32_SPEC_CTRL for SSBD.
217  */
218 void
219 x86_virt_spec_ctrl(u64 guest_virt_spec_ctrl, bool setguest)
220 {
221 	u64 guestval, hostval;
222 	struct thread_info *ti = current_thread_info();
223 
224 	/*
225 	 * If SSBD is not handled in MSR_SPEC_CTRL on AMD, update
226 	 * MSR_AMD64_L2_CFG or MSR_VIRT_SPEC_CTRL if supported.
227 	 */
228 	if (!static_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
229 	    !static_cpu_has(X86_FEATURE_VIRT_SSBD))
230 		return;
231 
232 	/*
233 	 * If the host has SSBD mitigation enabled, force it in the host's
234 	 * virtual MSR value. If its not permanently enabled, evaluate
235 	 * current's TIF_SSBD thread flag.
236 	 */
237 	if (static_cpu_has(X86_FEATURE_SPEC_STORE_BYPASS_DISABLE))
238 		hostval = SPEC_CTRL_SSBD;
239 	else
240 		hostval = ssbd_tif_to_spec_ctrl(ti->flags);
241 
242 	/* Sanitize the guest value */
243 	guestval = guest_virt_spec_ctrl & SPEC_CTRL_SSBD;
244 
245 	if (hostval != guestval) {
246 		unsigned long tif;
247 
248 		tif = setguest ? ssbd_spec_ctrl_to_tif(guestval) :
249 				 ssbd_spec_ctrl_to_tif(hostval);
250 
251 		speculation_ctrl_update(tif);
252 	}
253 }
254 EXPORT_SYMBOL_GPL(x86_virt_spec_ctrl);
255 
256 static void x86_amd_ssb_disable(void)
257 {
258 	u64 msrval = x86_amd_ls_cfg_base | x86_amd_ls_cfg_ssbd_mask;
259 
260 	if (boot_cpu_has(X86_FEATURE_VIRT_SSBD))
261 		wrmsrl(MSR_AMD64_VIRT_SPEC_CTRL, SPEC_CTRL_SSBD);
262 	else if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD))
263 		wrmsrl(MSR_AMD64_LS_CFG, msrval);
264 }
265 
266 #undef pr_fmt
267 #define pr_fmt(fmt)	"MDS: " fmt
268 
269 /* Default mitigation for MDS-affected CPUs */
270 static enum mds_mitigations mds_mitigation __ro_after_init = MDS_MITIGATION_FULL;
271 static bool mds_nosmt __ro_after_init = false;
272 
273 static const char * const mds_strings[] = {
274 	[MDS_MITIGATION_OFF]	= "Vulnerable",
275 	[MDS_MITIGATION_FULL]	= "Mitigation: Clear CPU buffers",
276 	[MDS_MITIGATION_VMWERV]	= "Vulnerable: Clear CPU buffers attempted, no microcode",
277 };
278 
279 static void __init mds_select_mitigation(void)
280 {
281 	if (!boot_cpu_has_bug(X86_BUG_MDS) || cpu_mitigations_off()) {
282 		mds_mitigation = MDS_MITIGATION_OFF;
283 		return;
284 	}
285 
286 	if (mds_mitigation == MDS_MITIGATION_FULL) {
287 		if (!boot_cpu_has(X86_FEATURE_MD_CLEAR))
288 			mds_mitigation = MDS_MITIGATION_VMWERV;
289 
290 		static_branch_enable(&mds_user_clear);
291 
292 		if (!boot_cpu_has(X86_BUG_MSBDS_ONLY) &&
293 		    (mds_nosmt || cpu_mitigations_auto_nosmt()))
294 			cpu_smt_disable(false);
295 	}
296 }
297 
298 static int __init mds_cmdline(char *str)
299 {
300 	if (!boot_cpu_has_bug(X86_BUG_MDS))
301 		return 0;
302 
303 	if (!str)
304 		return -EINVAL;
305 
306 	if (!strcmp(str, "off"))
307 		mds_mitigation = MDS_MITIGATION_OFF;
308 	else if (!strcmp(str, "full"))
309 		mds_mitigation = MDS_MITIGATION_FULL;
310 	else if (!strcmp(str, "full,nosmt")) {
311 		mds_mitigation = MDS_MITIGATION_FULL;
312 		mds_nosmt = true;
313 	}
314 
315 	return 0;
316 }
317 early_param("mds", mds_cmdline);
318 
319 #undef pr_fmt
320 #define pr_fmt(fmt)	"TAA: " fmt
321 
322 enum taa_mitigations {
323 	TAA_MITIGATION_OFF,
324 	TAA_MITIGATION_UCODE_NEEDED,
325 	TAA_MITIGATION_VERW,
326 	TAA_MITIGATION_TSX_DISABLED,
327 };
328 
329 /* Default mitigation for TAA-affected CPUs */
330 static enum taa_mitigations taa_mitigation __ro_after_init = TAA_MITIGATION_VERW;
331 static bool taa_nosmt __ro_after_init;
332 
333 static const char * const taa_strings[] = {
334 	[TAA_MITIGATION_OFF]		= "Vulnerable",
335 	[TAA_MITIGATION_UCODE_NEEDED]	= "Vulnerable: Clear CPU buffers attempted, no microcode",
336 	[TAA_MITIGATION_VERW]		= "Mitigation: Clear CPU buffers",
337 	[TAA_MITIGATION_TSX_DISABLED]	= "Mitigation: TSX disabled",
338 };
339 
340 static void __init taa_select_mitigation(void)
341 {
342 	u64 ia32_cap;
343 
344 	if (!boot_cpu_has_bug(X86_BUG_TAA)) {
345 		taa_mitigation = TAA_MITIGATION_OFF;
346 		return;
347 	}
348 
349 	/* TSX previously disabled by tsx=off */
350 	if (!boot_cpu_has(X86_FEATURE_RTM)) {
351 		taa_mitigation = TAA_MITIGATION_TSX_DISABLED;
352 		return;
353 	}
354 
355 	if (cpu_mitigations_off()) {
356 		taa_mitigation = TAA_MITIGATION_OFF;
357 		return;
358 	}
359 
360 	/*
361 	 * TAA mitigation via VERW is turned off if both
362 	 * tsx_async_abort=off and mds=off are specified.
363 	 */
364 	if (taa_mitigation == TAA_MITIGATION_OFF &&
365 	    mds_mitigation == MDS_MITIGATION_OFF)
366 		return;
367 
368 	if (boot_cpu_has(X86_FEATURE_MD_CLEAR))
369 		taa_mitigation = TAA_MITIGATION_VERW;
370 	else
371 		taa_mitigation = TAA_MITIGATION_UCODE_NEEDED;
372 
373 	/*
374 	 * VERW doesn't clear the CPU buffers when MD_CLEAR=1 and MDS_NO=1.
375 	 * A microcode update fixes this behavior to clear CPU buffers. It also
376 	 * adds support for MSR_IA32_TSX_CTRL which is enumerated by the
377 	 * ARCH_CAP_TSX_CTRL_MSR bit.
378 	 *
379 	 * On MDS_NO=1 CPUs if ARCH_CAP_TSX_CTRL_MSR is not set, microcode
380 	 * update is required.
381 	 */
382 	ia32_cap = x86_read_arch_cap_msr();
383 	if ( (ia32_cap & ARCH_CAP_MDS_NO) &&
384 	    !(ia32_cap & ARCH_CAP_TSX_CTRL_MSR))
385 		taa_mitigation = TAA_MITIGATION_UCODE_NEEDED;
386 
387 	/*
388 	 * TSX is enabled, select alternate mitigation for TAA which is
389 	 * the same as MDS. Enable MDS static branch to clear CPU buffers.
390 	 *
391 	 * For guests that can't determine whether the correct microcode is
392 	 * present on host, enable the mitigation for UCODE_NEEDED as well.
393 	 */
394 	static_branch_enable(&mds_user_clear);
395 
396 	if (taa_nosmt || cpu_mitigations_auto_nosmt())
397 		cpu_smt_disable(false);
398 }
399 
400 static int __init tsx_async_abort_parse_cmdline(char *str)
401 {
402 	if (!boot_cpu_has_bug(X86_BUG_TAA))
403 		return 0;
404 
405 	if (!str)
406 		return -EINVAL;
407 
408 	if (!strcmp(str, "off")) {
409 		taa_mitigation = TAA_MITIGATION_OFF;
410 	} else if (!strcmp(str, "full")) {
411 		taa_mitigation = TAA_MITIGATION_VERW;
412 	} else if (!strcmp(str, "full,nosmt")) {
413 		taa_mitigation = TAA_MITIGATION_VERW;
414 		taa_nosmt = true;
415 	}
416 
417 	return 0;
418 }
419 early_param("tsx_async_abort", tsx_async_abort_parse_cmdline);
420 
421 #undef pr_fmt
422 #define pr_fmt(fmt)	"MMIO Stale Data: " fmt
423 
424 enum mmio_mitigations {
425 	MMIO_MITIGATION_OFF,
426 	MMIO_MITIGATION_UCODE_NEEDED,
427 	MMIO_MITIGATION_VERW,
428 };
429 
430 /* Default mitigation for Processor MMIO Stale Data vulnerabilities */
431 static enum mmio_mitigations mmio_mitigation __ro_after_init = MMIO_MITIGATION_VERW;
432 static bool mmio_nosmt __ro_after_init = false;
433 
434 static const char * const mmio_strings[] = {
435 	[MMIO_MITIGATION_OFF]		= "Vulnerable",
436 	[MMIO_MITIGATION_UCODE_NEEDED]	= "Vulnerable: Clear CPU buffers attempted, no microcode",
437 	[MMIO_MITIGATION_VERW]		= "Mitigation: Clear CPU buffers",
438 };
439 
440 static void __init mmio_select_mitigation(void)
441 {
442 	u64 ia32_cap;
443 
444 	if (!boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA) ||
445 	     boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN) ||
446 	     cpu_mitigations_off()) {
447 		mmio_mitigation = MMIO_MITIGATION_OFF;
448 		return;
449 	}
450 
451 	if (mmio_mitigation == MMIO_MITIGATION_OFF)
452 		return;
453 
454 	ia32_cap = x86_read_arch_cap_msr();
455 
456 	/*
457 	 * Enable CPU buffer clear mitigation for host and VMM, if also affected
458 	 * by MDS or TAA. Otherwise, enable mitigation for VMM only.
459 	 */
460 	if (boot_cpu_has_bug(X86_BUG_MDS) || (boot_cpu_has_bug(X86_BUG_TAA) &&
461 					      boot_cpu_has(X86_FEATURE_RTM)))
462 		static_branch_enable(&mds_user_clear);
463 	else
464 		static_branch_enable(&mmio_stale_data_clear);
465 
466 	/*
467 	 * If Processor-MMIO-Stale-Data bug is present and Fill Buffer data can
468 	 * be propagated to uncore buffers, clearing the Fill buffers on idle
469 	 * is required irrespective of SMT state.
470 	 */
471 	if (!(ia32_cap & ARCH_CAP_FBSDP_NO))
472 		static_branch_enable(&mds_idle_clear);
473 
474 	/*
475 	 * Check if the system has the right microcode.
476 	 *
477 	 * CPU Fill buffer clear mitigation is enumerated by either an explicit
478 	 * FB_CLEAR or by the presence of both MD_CLEAR and L1D_FLUSH on MDS
479 	 * affected systems.
480 	 */
481 	if ((ia32_cap & ARCH_CAP_FB_CLEAR) ||
482 	    (boot_cpu_has(X86_FEATURE_MD_CLEAR) &&
483 	     boot_cpu_has(X86_FEATURE_FLUSH_L1D) &&
484 	     !(ia32_cap & ARCH_CAP_MDS_NO)))
485 		mmio_mitigation = MMIO_MITIGATION_VERW;
486 	else
487 		mmio_mitigation = MMIO_MITIGATION_UCODE_NEEDED;
488 
489 	if (mmio_nosmt || cpu_mitigations_auto_nosmt())
490 		cpu_smt_disable(false);
491 }
492 
493 static int __init mmio_stale_data_parse_cmdline(char *str)
494 {
495 	if (!boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA))
496 		return 0;
497 
498 	if (!str)
499 		return -EINVAL;
500 
501 	if (!strcmp(str, "off")) {
502 		mmio_mitigation = MMIO_MITIGATION_OFF;
503 	} else if (!strcmp(str, "full")) {
504 		mmio_mitigation = MMIO_MITIGATION_VERW;
505 	} else if (!strcmp(str, "full,nosmt")) {
506 		mmio_mitigation = MMIO_MITIGATION_VERW;
507 		mmio_nosmt = true;
508 	}
509 
510 	return 0;
511 }
512 early_param("mmio_stale_data", mmio_stale_data_parse_cmdline);
513 
514 #undef pr_fmt
515 #define pr_fmt(fmt)     "" fmt
516 
517 static void __init md_clear_update_mitigation(void)
518 {
519 	if (cpu_mitigations_off())
520 		return;
521 
522 	if (!static_key_enabled(&mds_user_clear))
523 		goto out;
524 
525 	/*
526 	 * mds_user_clear is now enabled. Update MDS, TAA and MMIO Stale Data
527 	 * mitigation, if necessary.
528 	 */
529 	if (mds_mitigation == MDS_MITIGATION_OFF &&
530 	    boot_cpu_has_bug(X86_BUG_MDS)) {
531 		mds_mitigation = MDS_MITIGATION_FULL;
532 		mds_select_mitigation();
533 	}
534 	if (taa_mitigation == TAA_MITIGATION_OFF &&
535 	    boot_cpu_has_bug(X86_BUG_TAA)) {
536 		taa_mitigation = TAA_MITIGATION_VERW;
537 		taa_select_mitigation();
538 	}
539 	if (mmio_mitigation == MMIO_MITIGATION_OFF &&
540 	    boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA)) {
541 		mmio_mitigation = MMIO_MITIGATION_VERW;
542 		mmio_select_mitigation();
543 	}
544 out:
545 	if (boot_cpu_has_bug(X86_BUG_MDS))
546 		pr_info("MDS: %s\n", mds_strings[mds_mitigation]);
547 	if (boot_cpu_has_bug(X86_BUG_TAA))
548 		pr_info("TAA: %s\n", taa_strings[taa_mitigation]);
549 	if (boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA))
550 		pr_info("MMIO Stale Data: %s\n", mmio_strings[mmio_mitigation]);
551 	else if (boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN))
552 		pr_info("MMIO Stale Data: Unknown: No mitigations\n");
553 }
554 
555 static void __init md_clear_select_mitigation(void)
556 {
557 	mds_select_mitigation();
558 	taa_select_mitigation();
559 	mmio_select_mitigation();
560 
561 	/*
562 	 * As MDS, TAA and MMIO Stale Data mitigations are inter-related, update
563 	 * and print their mitigation after MDS, TAA and MMIO Stale Data
564 	 * mitigation selection is done.
565 	 */
566 	md_clear_update_mitigation();
567 }
568 
569 #undef pr_fmt
570 #define pr_fmt(fmt)	"SRBDS: " fmt
571 
572 enum srbds_mitigations {
573 	SRBDS_MITIGATION_OFF,
574 	SRBDS_MITIGATION_UCODE_NEEDED,
575 	SRBDS_MITIGATION_FULL,
576 	SRBDS_MITIGATION_TSX_OFF,
577 	SRBDS_MITIGATION_HYPERVISOR,
578 };
579 
580 static enum srbds_mitigations srbds_mitigation __ro_after_init = SRBDS_MITIGATION_FULL;
581 
582 static const char * const srbds_strings[] = {
583 	[SRBDS_MITIGATION_OFF]		= "Vulnerable",
584 	[SRBDS_MITIGATION_UCODE_NEEDED]	= "Vulnerable: No microcode",
585 	[SRBDS_MITIGATION_FULL]		= "Mitigation: Microcode",
586 	[SRBDS_MITIGATION_TSX_OFF]	= "Mitigation: TSX disabled",
587 	[SRBDS_MITIGATION_HYPERVISOR]	= "Unknown: Dependent on hypervisor status",
588 };
589 
590 static bool srbds_off;
591 
592 void update_srbds_msr(void)
593 {
594 	u64 mcu_ctrl;
595 
596 	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
597 		return;
598 
599 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
600 		return;
601 
602 	if (srbds_mitigation == SRBDS_MITIGATION_UCODE_NEEDED)
603 		return;
604 
605 	/*
606 	 * A MDS_NO CPU for which SRBDS mitigation is not needed due to TSX
607 	 * being disabled and it hasn't received the SRBDS MSR microcode.
608 	 */
609 	if (!boot_cpu_has(X86_FEATURE_SRBDS_CTRL))
610 		return;
611 
612 	rdmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
613 
614 	switch (srbds_mitigation) {
615 	case SRBDS_MITIGATION_OFF:
616 	case SRBDS_MITIGATION_TSX_OFF:
617 		mcu_ctrl |= RNGDS_MITG_DIS;
618 		break;
619 	case SRBDS_MITIGATION_FULL:
620 		mcu_ctrl &= ~RNGDS_MITG_DIS;
621 		break;
622 	default:
623 		break;
624 	}
625 
626 	wrmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
627 }
628 
629 static void __init srbds_select_mitigation(void)
630 {
631 	u64 ia32_cap;
632 
633 	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
634 		return;
635 
636 	/*
637 	 * Check to see if this is one of the MDS_NO systems supporting TSX that
638 	 * are only exposed to SRBDS when TSX is enabled or when CPU is affected
639 	 * by Processor MMIO Stale Data vulnerability.
640 	 */
641 	ia32_cap = x86_read_arch_cap_msr();
642 	if ((ia32_cap & ARCH_CAP_MDS_NO) && !boot_cpu_has(X86_FEATURE_RTM) &&
643 	    !boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA))
644 		srbds_mitigation = SRBDS_MITIGATION_TSX_OFF;
645 	else if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
646 		srbds_mitigation = SRBDS_MITIGATION_HYPERVISOR;
647 	else if (!boot_cpu_has(X86_FEATURE_SRBDS_CTRL))
648 		srbds_mitigation = SRBDS_MITIGATION_UCODE_NEEDED;
649 	else if (cpu_mitigations_off() || srbds_off)
650 		srbds_mitigation = SRBDS_MITIGATION_OFF;
651 
652 	update_srbds_msr();
653 	pr_info("%s\n", srbds_strings[srbds_mitigation]);
654 }
655 
656 static int __init srbds_parse_cmdline(char *str)
657 {
658 	if (!str)
659 		return -EINVAL;
660 
661 	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
662 		return 0;
663 
664 	srbds_off = !strcmp(str, "off");
665 	return 0;
666 }
667 early_param("srbds", srbds_parse_cmdline);
668 
669 #undef pr_fmt
670 #define pr_fmt(fmt)     "L1D Flush : " fmt
671 
672 enum l1d_flush_mitigations {
673 	L1D_FLUSH_OFF = 0,
674 	L1D_FLUSH_ON,
675 };
676 
677 static enum l1d_flush_mitigations l1d_flush_mitigation __initdata = L1D_FLUSH_OFF;
678 
679 static void __init l1d_flush_select_mitigation(void)
680 {
681 	if (!l1d_flush_mitigation || !boot_cpu_has(X86_FEATURE_FLUSH_L1D))
682 		return;
683 
684 	static_branch_enable(&switch_mm_cond_l1d_flush);
685 	pr_info("Conditional flush on switch_mm() enabled\n");
686 }
687 
688 static int __init l1d_flush_parse_cmdline(char *str)
689 {
690 	if (!strcmp(str, "on"))
691 		l1d_flush_mitigation = L1D_FLUSH_ON;
692 
693 	return 0;
694 }
695 early_param("l1d_flush", l1d_flush_parse_cmdline);
696 
697 #undef pr_fmt
698 #define pr_fmt(fmt)     "Spectre V1 : " fmt
699 
700 enum spectre_v1_mitigation {
701 	SPECTRE_V1_MITIGATION_NONE,
702 	SPECTRE_V1_MITIGATION_AUTO,
703 };
704 
705 static enum spectre_v1_mitigation spectre_v1_mitigation __ro_after_init =
706 	SPECTRE_V1_MITIGATION_AUTO;
707 
708 static const char * const spectre_v1_strings[] = {
709 	[SPECTRE_V1_MITIGATION_NONE] = "Vulnerable: __user pointer sanitization and usercopy barriers only; no swapgs barriers",
710 	[SPECTRE_V1_MITIGATION_AUTO] = "Mitigation: usercopy/swapgs barriers and __user pointer sanitization",
711 };
712 
713 /*
714  * Does SMAP provide full mitigation against speculative kernel access to
715  * userspace?
716  */
717 static bool smap_works_speculatively(void)
718 {
719 	if (!boot_cpu_has(X86_FEATURE_SMAP))
720 		return false;
721 
722 	/*
723 	 * On CPUs which are vulnerable to Meltdown, SMAP does not
724 	 * prevent speculative access to user data in the L1 cache.
725 	 * Consider SMAP to be non-functional as a mitigation on these
726 	 * CPUs.
727 	 */
728 	if (boot_cpu_has(X86_BUG_CPU_MELTDOWN))
729 		return false;
730 
731 	return true;
732 }
733 
734 static void __init spectre_v1_select_mitigation(void)
735 {
736 	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V1) || cpu_mitigations_off()) {
737 		spectre_v1_mitigation = SPECTRE_V1_MITIGATION_NONE;
738 		return;
739 	}
740 
741 	if (spectre_v1_mitigation == SPECTRE_V1_MITIGATION_AUTO) {
742 		/*
743 		 * With Spectre v1, a user can speculatively control either
744 		 * path of a conditional swapgs with a user-controlled GS
745 		 * value.  The mitigation is to add lfences to both code paths.
746 		 *
747 		 * If FSGSBASE is enabled, the user can put a kernel address in
748 		 * GS, in which case SMAP provides no protection.
749 		 *
750 		 * If FSGSBASE is disabled, the user can only put a user space
751 		 * address in GS.  That makes an attack harder, but still
752 		 * possible if there's no SMAP protection.
753 		 */
754 		if (boot_cpu_has(X86_FEATURE_FSGSBASE) ||
755 		    !smap_works_speculatively()) {
756 			/*
757 			 * Mitigation can be provided from SWAPGS itself or
758 			 * PTI as the CR3 write in the Meltdown mitigation
759 			 * is serializing.
760 			 *
761 			 * If neither is there, mitigate with an LFENCE to
762 			 * stop speculation through swapgs.
763 			 */
764 			if (boot_cpu_has_bug(X86_BUG_SWAPGS) &&
765 			    !boot_cpu_has(X86_FEATURE_PTI))
766 				setup_force_cpu_cap(X86_FEATURE_FENCE_SWAPGS_USER);
767 
768 			/*
769 			 * Enable lfences in the kernel entry (non-swapgs)
770 			 * paths, to prevent user entry from speculatively
771 			 * skipping swapgs.
772 			 */
773 			setup_force_cpu_cap(X86_FEATURE_FENCE_SWAPGS_KERNEL);
774 		}
775 	}
776 
777 	pr_info("%s\n", spectre_v1_strings[spectre_v1_mitigation]);
778 }
779 
780 static int __init nospectre_v1_cmdline(char *str)
781 {
782 	spectre_v1_mitigation = SPECTRE_V1_MITIGATION_NONE;
783 	return 0;
784 }
785 early_param("nospectre_v1", nospectre_v1_cmdline);
786 
787 enum spectre_v2_mitigation spectre_v2_enabled __ro_after_init = SPECTRE_V2_NONE;
788 
789 #undef pr_fmt
790 #define pr_fmt(fmt)     "RETBleed: " fmt
791 
792 enum retbleed_mitigation {
793 	RETBLEED_MITIGATION_NONE,
794 	RETBLEED_MITIGATION_UNRET,
795 	RETBLEED_MITIGATION_IBPB,
796 	RETBLEED_MITIGATION_IBRS,
797 	RETBLEED_MITIGATION_EIBRS,
798 	RETBLEED_MITIGATION_STUFF,
799 };
800 
801 enum retbleed_mitigation_cmd {
802 	RETBLEED_CMD_OFF,
803 	RETBLEED_CMD_AUTO,
804 	RETBLEED_CMD_UNRET,
805 	RETBLEED_CMD_IBPB,
806 	RETBLEED_CMD_STUFF,
807 };
808 
809 static const char * const retbleed_strings[] = {
810 	[RETBLEED_MITIGATION_NONE]	= "Vulnerable",
811 	[RETBLEED_MITIGATION_UNRET]	= "Mitigation: untrained return thunk",
812 	[RETBLEED_MITIGATION_IBPB]	= "Mitigation: IBPB",
813 	[RETBLEED_MITIGATION_IBRS]	= "Mitigation: IBRS",
814 	[RETBLEED_MITIGATION_EIBRS]	= "Mitigation: Enhanced IBRS",
815 	[RETBLEED_MITIGATION_STUFF]	= "Mitigation: Stuffing",
816 };
817 
818 static enum retbleed_mitigation retbleed_mitigation __ro_after_init =
819 	RETBLEED_MITIGATION_NONE;
820 static enum retbleed_mitigation_cmd retbleed_cmd __ro_after_init =
821 	RETBLEED_CMD_AUTO;
822 
823 static int __ro_after_init retbleed_nosmt = false;
824 
825 static int __init retbleed_parse_cmdline(char *str)
826 {
827 	if (!str)
828 		return -EINVAL;
829 
830 	while (str) {
831 		char *next = strchr(str, ',');
832 		if (next) {
833 			*next = 0;
834 			next++;
835 		}
836 
837 		if (!strcmp(str, "off")) {
838 			retbleed_cmd = RETBLEED_CMD_OFF;
839 		} else if (!strcmp(str, "auto")) {
840 			retbleed_cmd = RETBLEED_CMD_AUTO;
841 		} else if (!strcmp(str, "unret")) {
842 			retbleed_cmd = RETBLEED_CMD_UNRET;
843 		} else if (!strcmp(str, "ibpb")) {
844 			retbleed_cmd = RETBLEED_CMD_IBPB;
845 		} else if (!strcmp(str, "stuff")) {
846 			retbleed_cmd = RETBLEED_CMD_STUFF;
847 		} else if (!strcmp(str, "nosmt")) {
848 			retbleed_nosmt = true;
849 		} else if (!strcmp(str, "force")) {
850 			setup_force_cpu_bug(X86_BUG_RETBLEED);
851 		} else {
852 			pr_err("Ignoring unknown retbleed option (%s).", str);
853 		}
854 
855 		str = next;
856 	}
857 
858 	return 0;
859 }
860 early_param("retbleed", retbleed_parse_cmdline);
861 
862 #define RETBLEED_UNTRAIN_MSG "WARNING: BTB untrained return thunk mitigation is only effective on AMD/Hygon!\n"
863 #define RETBLEED_INTEL_MSG "WARNING: Spectre v2 mitigation leaves CPU vulnerable to RETBleed attacks, data leaks possible!\n"
864 
865 static void __init retbleed_select_mitigation(void)
866 {
867 	bool mitigate_smt = false;
868 
869 	if (!boot_cpu_has_bug(X86_BUG_RETBLEED) || cpu_mitigations_off())
870 		return;
871 
872 	switch (retbleed_cmd) {
873 	case RETBLEED_CMD_OFF:
874 		return;
875 
876 	case RETBLEED_CMD_UNRET:
877 		if (IS_ENABLED(CONFIG_CPU_UNRET_ENTRY)) {
878 			retbleed_mitigation = RETBLEED_MITIGATION_UNRET;
879 		} else {
880 			pr_err("WARNING: kernel not compiled with CPU_UNRET_ENTRY.\n");
881 			goto do_cmd_auto;
882 		}
883 		break;
884 
885 	case RETBLEED_CMD_IBPB:
886 		if (!boot_cpu_has(X86_FEATURE_IBPB)) {
887 			pr_err("WARNING: CPU does not support IBPB.\n");
888 			goto do_cmd_auto;
889 		} else if (IS_ENABLED(CONFIG_CPU_IBPB_ENTRY)) {
890 			retbleed_mitigation = RETBLEED_MITIGATION_IBPB;
891 		} else {
892 			pr_err("WARNING: kernel not compiled with CPU_IBPB_ENTRY.\n");
893 			goto do_cmd_auto;
894 		}
895 		break;
896 
897 	case RETBLEED_CMD_STUFF:
898 		if (IS_ENABLED(CONFIG_CALL_DEPTH_TRACKING) &&
899 		    spectre_v2_enabled == SPECTRE_V2_RETPOLINE) {
900 			retbleed_mitigation = RETBLEED_MITIGATION_STUFF;
901 
902 		} else {
903 			if (IS_ENABLED(CONFIG_CALL_DEPTH_TRACKING))
904 				pr_err("WARNING: retbleed=stuff depends on spectre_v2=retpoline\n");
905 			else
906 				pr_err("WARNING: kernel not compiled with CALL_DEPTH_TRACKING.\n");
907 
908 			goto do_cmd_auto;
909 		}
910 		break;
911 
912 do_cmd_auto:
913 	case RETBLEED_CMD_AUTO:
914 	default:
915 		if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
916 		    boot_cpu_data.x86_vendor == X86_VENDOR_HYGON) {
917 			if (IS_ENABLED(CONFIG_CPU_UNRET_ENTRY))
918 				retbleed_mitigation = RETBLEED_MITIGATION_UNRET;
919 			else if (IS_ENABLED(CONFIG_CPU_IBPB_ENTRY) && boot_cpu_has(X86_FEATURE_IBPB))
920 				retbleed_mitigation = RETBLEED_MITIGATION_IBPB;
921 		}
922 
923 		/*
924 		 * The Intel mitigation (IBRS or eIBRS) was already selected in
925 		 * spectre_v2_select_mitigation().  'retbleed_mitigation' will
926 		 * be set accordingly below.
927 		 */
928 
929 		break;
930 	}
931 
932 	switch (retbleed_mitigation) {
933 	case RETBLEED_MITIGATION_UNRET:
934 		setup_force_cpu_cap(X86_FEATURE_RETHUNK);
935 		setup_force_cpu_cap(X86_FEATURE_UNRET);
936 
937 		if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD &&
938 		    boot_cpu_data.x86_vendor != X86_VENDOR_HYGON)
939 			pr_err(RETBLEED_UNTRAIN_MSG);
940 
941 		mitigate_smt = true;
942 		break;
943 
944 	case RETBLEED_MITIGATION_IBPB:
945 		setup_force_cpu_cap(X86_FEATURE_ENTRY_IBPB);
946 		mitigate_smt = true;
947 		break;
948 
949 	case RETBLEED_MITIGATION_STUFF:
950 		setup_force_cpu_cap(X86_FEATURE_RETHUNK);
951 		setup_force_cpu_cap(X86_FEATURE_CALL_DEPTH);
952 		x86_set_skl_return_thunk();
953 		break;
954 
955 	default:
956 		break;
957 	}
958 
959 	if (mitigate_smt && !boot_cpu_has(X86_FEATURE_STIBP) &&
960 	    (retbleed_nosmt || cpu_mitigations_auto_nosmt()))
961 		cpu_smt_disable(false);
962 
963 	/*
964 	 * Let IBRS trump all on Intel without affecting the effects of the
965 	 * retbleed= cmdline option except for call depth based stuffing
966 	 */
967 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) {
968 		switch (spectre_v2_enabled) {
969 		case SPECTRE_V2_IBRS:
970 			retbleed_mitigation = RETBLEED_MITIGATION_IBRS;
971 			break;
972 		case SPECTRE_V2_EIBRS:
973 		case SPECTRE_V2_EIBRS_RETPOLINE:
974 		case SPECTRE_V2_EIBRS_LFENCE:
975 			retbleed_mitigation = RETBLEED_MITIGATION_EIBRS;
976 			break;
977 		default:
978 			if (retbleed_mitigation != RETBLEED_MITIGATION_STUFF)
979 				pr_err(RETBLEED_INTEL_MSG);
980 		}
981 	}
982 
983 	pr_info("%s\n", retbleed_strings[retbleed_mitigation]);
984 }
985 
986 #undef pr_fmt
987 #define pr_fmt(fmt)     "Spectre V2 : " fmt
988 
989 static enum spectre_v2_user_mitigation spectre_v2_user_stibp __ro_after_init =
990 	SPECTRE_V2_USER_NONE;
991 static enum spectre_v2_user_mitigation spectre_v2_user_ibpb __ro_after_init =
992 	SPECTRE_V2_USER_NONE;
993 
994 #ifdef CONFIG_RETPOLINE
995 static bool spectre_v2_bad_module;
996 
997 bool retpoline_module_ok(bool has_retpoline)
998 {
999 	if (spectre_v2_enabled == SPECTRE_V2_NONE || has_retpoline)
1000 		return true;
1001 
1002 	pr_err("System may be vulnerable to spectre v2\n");
1003 	spectre_v2_bad_module = true;
1004 	return false;
1005 }
1006 
1007 static inline const char *spectre_v2_module_string(void)
1008 {
1009 	return spectre_v2_bad_module ? " - vulnerable module loaded" : "";
1010 }
1011 #else
1012 static inline const char *spectre_v2_module_string(void) { return ""; }
1013 #endif
1014 
1015 #define SPECTRE_V2_LFENCE_MSG "WARNING: LFENCE mitigation is not recommended for this CPU, data leaks possible!\n"
1016 #define SPECTRE_V2_EIBRS_EBPF_MSG "WARNING: Unprivileged eBPF is enabled with eIBRS on, data leaks possible via Spectre v2 BHB attacks!\n"
1017 #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"
1018 #define SPECTRE_V2_IBRS_PERF_MSG "WARNING: IBRS mitigation selected on Enhanced IBRS CPU, this may cause unnecessary performance loss\n"
1019 
1020 #ifdef CONFIG_BPF_SYSCALL
1021 void unpriv_ebpf_notify(int new_state)
1022 {
1023 	if (new_state)
1024 		return;
1025 
1026 	/* Unprivileged eBPF is enabled */
1027 
1028 	switch (spectre_v2_enabled) {
1029 	case SPECTRE_V2_EIBRS:
1030 		pr_err(SPECTRE_V2_EIBRS_EBPF_MSG);
1031 		break;
1032 	case SPECTRE_V2_EIBRS_LFENCE:
1033 		if (sched_smt_active())
1034 			pr_err(SPECTRE_V2_EIBRS_LFENCE_EBPF_SMT_MSG);
1035 		break;
1036 	default:
1037 		break;
1038 	}
1039 }
1040 #endif
1041 
1042 static inline bool match_option(const char *arg, int arglen, const char *opt)
1043 {
1044 	int len = strlen(opt);
1045 
1046 	return len == arglen && !strncmp(arg, opt, len);
1047 }
1048 
1049 /* The kernel command line selection for spectre v2 */
1050 enum spectre_v2_mitigation_cmd {
1051 	SPECTRE_V2_CMD_NONE,
1052 	SPECTRE_V2_CMD_AUTO,
1053 	SPECTRE_V2_CMD_FORCE,
1054 	SPECTRE_V2_CMD_RETPOLINE,
1055 	SPECTRE_V2_CMD_RETPOLINE_GENERIC,
1056 	SPECTRE_V2_CMD_RETPOLINE_LFENCE,
1057 	SPECTRE_V2_CMD_EIBRS,
1058 	SPECTRE_V2_CMD_EIBRS_RETPOLINE,
1059 	SPECTRE_V2_CMD_EIBRS_LFENCE,
1060 	SPECTRE_V2_CMD_IBRS,
1061 };
1062 
1063 enum spectre_v2_user_cmd {
1064 	SPECTRE_V2_USER_CMD_NONE,
1065 	SPECTRE_V2_USER_CMD_AUTO,
1066 	SPECTRE_V2_USER_CMD_FORCE,
1067 	SPECTRE_V2_USER_CMD_PRCTL,
1068 	SPECTRE_V2_USER_CMD_PRCTL_IBPB,
1069 	SPECTRE_V2_USER_CMD_SECCOMP,
1070 	SPECTRE_V2_USER_CMD_SECCOMP_IBPB,
1071 };
1072 
1073 static const char * const spectre_v2_user_strings[] = {
1074 	[SPECTRE_V2_USER_NONE]			= "User space: Vulnerable",
1075 	[SPECTRE_V2_USER_STRICT]		= "User space: Mitigation: STIBP protection",
1076 	[SPECTRE_V2_USER_STRICT_PREFERRED]	= "User space: Mitigation: STIBP always-on protection",
1077 	[SPECTRE_V2_USER_PRCTL]			= "User space: Mitigation: STIBP via prctl",
1078 	[SPECTRE_V2_USER_SECCOMP]		= "User space: Mitigation: STIBP via seccomp and prctl",
1079 };
1080 
1081 static const struct {
1082 	const char			*option;
1083 	enum spectre_v2_user_cmd	cmd;
1084 	bool				secure;
1085 } v2_user_options[] __initconst = {
1086 	{ "auto",		SPECTRE_V2_USER_CMD_AUTO,		false },
1087 	{ "off",		SPECTRE_V2_USER_CMD_NONE,		false },
1088 	{ "on",			SPECTRE_V2_USER_CMD_FORCE,		true  },
1089 	{ "prctl",		SPECTRE_V2_USER_CMD_PRCTL,		false },
1090 	{ "prctl,ibpb",		SPECTRE_V2_USER_CMD_PRCTL_IBPB,		false },
1091 	{ "seccomp",		SPECTRE_V2_USER_CMD_SECCOMP,		false },
1092 	{ "seccomp,ibpb",	SPECTRE_V2_USER_CMD_SECCOMP_IBPB,	false },
1093 };
1094 
1095 static void __init spec_v2_user_print_cond(const char *reason, bool secure)
1096 {
1097 	if (boot_cpu_has_bug(X86_BUG_SPECTRE_V2) != secure)
1098 		pr_info("spectre_v2_user=%s forced on command line.\n", reason);
1099 }
1100 
1101 static __ro_after_init enum spectre_v2_mitigation_cmd spectre_v2_cmd;
1102 
1103 static enum spectre_v2_user_cmd __init
1104 spectre_v2_parse_user_cmdline(void)
1105 {
1106 	char arg[20];
1107 	int ret, i;
1108 
1109 	switch (spectre_v2_cmd) {
1110 	case SPECTRE_V2_CMD_NONE:
1111 		return SPECTRE_V2_USER_CMD_NONE;
1112 	case SPECTRE_V2_CMD_FORCE:
1113 		return SPECTRE_V2_USER_CMD_FORCE;
1114 	default:
1115 		break;
1116 	}
1117 
1118 	ret = cmdline_find_option(boot_command_line, "spectre_v2_user",
1119 				  arg, sizeof(arg));
1120 	if (ret < 0)
1121 		return SPECTRE_V2_USER_CMD_AUTO;
1122 
1123 	for (i = 0; i < ARRAY_SIZE(v2_user_options); i++) {
1124 		if (match_option(arg, ret, v2_user_options[i].option)) {
1125 			spec_v2_user_print_cond(v2_user_options[i].option,
1126 						v2_user_options[i].secure);
1127 			return v2_user_options[i].cmd;
1128 		}
1129 	}
1130 
1131 	pr_err("Unknown user space protection option (%s). Switching to AUTO select\n", arg);
1132 	return SPECTRE_V2_USER_CMD_AUTO;
1133 }
1134 
1135 static inline bool spectre_v2_in_ibrs_mode(enum spectre_v2_mitigation mode)
1136 {
1137 	return spectre_v2_in_eibrs_mode(mode) || mode == SPECTRE_V2_IBRS;
1138 }
1139 
1140 static void __init
1141 spectre_v2_user_select_mitigation(void)
1142 {
1143 	enum spectre_v2_user_mitigation mode = SPECTRE_V2_USER_NONE;
1144 	bool smt_possible = IS_ENABLED(CONFIG_SMP);
1145 	enum spectre_v2_user_cmd cmd;
1146 
1147 	if (!boot_cpu_has(X86_FEATURE_IBPB) && !boot_cpu_has(X86_FEATURE_STIBP))
1148 		return;
1149 
1150 	if (cpu_smt_control == CPU_SMT_FORCE_DISABLED ||
1151 	    cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
1152 		smt_possible = false;
1153 
1154 	cmd = spectre_v2_parse_user_cmdline();
1155 	switch (cmd) {
1156 	case SPECTRE_V2_USER_CMD_NONE:
1157 		goto set_mode;
1158 	case SPECTRE_V2_USER_CMD_FORCE:
1159 		mode = SPECTRE_V2_USER_STRICT;
1160 		break;
1161 	case SPECTRE_V2_USER_CMD_AUTO:
1162 	case SPECTRE_V2_USER_CMD_PRCTL:
1163 	case SPECTRE_V2_USER_CMD_PRCTL_IBPB:
1164 		mode = SPECTRE_V2_USER_PRCTL;
1165 		break;
1166 	case SPECTRE_V2_USER_CMD_SECCOMP:
1167 	case SPECTRE_V2_USER_CMD_SECCOMP_IBPB:
1168 		if (IS_ENABLED(CONFIG_SECCOMP))
1169 			mode = SPECTRE_V2_USER_SECCOMP;
1170 		else
1171 			mode = SPECTRE_V2_USER_PRCTL;
1172 		break;
1173 	}
1174 
1175 	/* Initialize Indirect Branch Prediction Barrier */
1176 	if (boot_cpu_has(X86_FEATURE_IBPB)) {
1177 		setup_force_cpu_cap(X86_FEATURE_USE_IBPB);
1178 
1179 		spectre_v2_user_ibpb = mode;
1180 		switch (cmd) {
1181 		case SPECTRE_V2_USER_CMD_FORCE:
1182 		case SPECTRE_V2_USER_CMD_PRCTL_IBPB:
1183 		case SPECTRE_V2_USER_CMD_SECCOMP_IBPB:
1184 			static_branch_enable(&switch_mm_always_ibpb);
1185 			spectre_v2_user_ibpb = SPECTRE_V2_USER_STRICT;
1186 			break;
1187 		case SPECTRE_V2_USER_CMD_PRCTL:
1188 		case SPECTRE_V2_USER_CMD_AUTO:
1189 		case SPECTRE_V2_USER_CMD_SECCOMP:
1190 			static_branch_enable(&switch_mm_cond_ibpb);
1191 			break;
1192 		default:
1193 			break;
1194 		}
1195 
1196 		pr_info("mitigation: Enabling %s Indirect Branch Prediction Barrier\n",
1197 			static_key_enabled(&switch_mm_always_ibpb) ?
1198 			"always-on" : "conditional");
1199 	}
1200 
1201 	/*
1202 	 * If no STIBP, enhanced IBRS is enabled, or SMT impossible, STIBP
1203 	 * is not required.
1204 	 *
1205 	 * Enhanced IBRS also protects against cross-thread branch target
1206 	 * injection in user-mode as the IBRS bit remains always set which
1207 	 * implicitly enables cross-thread protections.  However, in legacy IBRS
1208 	 * mode, the IBRS bit is set only on kernel entry and cleared on return
1209 	 * to userspace. This disables the implicit cross-thread protection,
1210 	 * so allow for STIBP to be selected in that case.
1211 	 */
1212 	if (!boot_cpu_has(X86_FEATURE_STIBP) ||
1213 	    !smt_possible ||
1214 	    spectre_v2_in_eibrs_mode(spectre_v2_enabled))
1215 		return;
1216 
1217 	/*
1218 	 * At this point, an STIBP mode other than "off" has been set.
1219 	 * If STIBP support is not being forced, check if STIBP always-on
1220 	 * is preferred.
1221 	 */
1222 	if (mode != SPECTRE_V2_USER_STRICT &&
1223 	    boot_cpu_has(X86_FEATURE_AMD_STIBP_ALWAYS_ON))
1224 		mode = SPECTRE_V2_USER_STRICT_PREFERRED;
1225 
1226 	if (retbleed_mitigation == RETBLEED_MITIGATION_UNRET ||
1227 	    retbleed_mitigation == RETBLEED_MITIGATION_IBPB) {
1228 		if (mode != SPECTRE_V2_USER_STRICT &&
1229 		    mode != SPECTRE_V2_USER_STRICT_PREFERRED)
1230 			pr_info("Selecting STIBP always-on mode to complement retbleed mitigation\n");
1231 		mode = SPECTRE_V2_USER_STRICT_PREFERRED;
1232 	}
1233 
1234 	spectre_v2_user_stibp = mode;
1235 
1236 set_mode:
1237 	pr_info("%s\n", spectre_v2_user_strings[mode]);
1238 }
1239 
1240 static const char * const spectre_v2_strings[] = {
1241 	[SPECTRE_V2_NONE]			= "Vulnerable",
1242 	[SPECTRE_V2_RETPOLINE]			= "Mitigation: Retpolines",
1243 	[SPECTRE_V2_LFENCE]			= "Mitigation: LFENCE",
1244 	[SPECTRE_V2_EIBRS]			= "Mitigation: Enhanced / Automatic IBRS",
1245 	[SPECTRE_V2_EIBRS_LFENCE]		= "Mitigation: Enhanced / Automatic IBRS + LFENCE",
1246 	[SPECTRE_V2_EIBRS_RETPOLINE]		= "Mitigation: Enhanced / Automatic IBRS + Retpolines",
1247 	[SPECTRE_V2_IBRS]			= "Mitigation: IBRS",
1248 };
1249 
1250 static const struct {
1251 	const char *option;
1252 	enum spectre_v2_mitigation_cmd cmd;
1253 	bool secure;
1254 } mitigation_options[] __initconst = {
1255 	{ "off",		SPECTRE_V2_CMD_NONE,		  false },
1256 	{ "on",			SPECTRE_V2_CMD_FORCE,		  true  },
1257 	{ "retpoline",		SPECTRE_V2_CMD_RETPOLINE,	  false },
1258 	{ "retpoline,amd",	SPECTRE_V2_CMD_RETPOLINE_LFENCE,  false },
1259 	{ "retpoline,lfence",	SPECTRE_V2_CMD_RETPOLINE_LFENCE,  false },
1260 	{ "retpoline,generic",	SPECTRE_V2_CMD_RETPOLINE_GENERIC, false },
1261 	{ "eibrs",		SPECTRE_V2_CMD_EIBRS,		  false },
1262 	{ "eibrs,lfence",	SPECTRE_V2_CMD_EIBRS_LFENCE,	  false },
1263 	{ "eibrs,retpoline",	SPECTRE_V2_CMD_EIBRS_RETPOLINE,	  false },
1264 	{ "auto",		SPECTRE_V2_CMD_AUTO,		  false },
1265 	{ "ibrs",		SPECTRE_V2_CMD_IBRS,              false },
1266 };
1267 
1268 static void __init spec_v2_print_cond(const char *reason, bool secure)
1269 {
1270 	if (boot_cpu_has_bug(X86_BUG_SPECTRE_V2) != secure)
1271 		pr_info("%s selected on command line.\n", reason);
1272 }
1273 
1274 static enum spectre_v2_mitigation_cmd __init spectre_v2_parse_cmdline(void)
1275 {
1276 	enum spectre_v2_mitigation_cmd cmd = SPECTRE_V2_CMD_AUTO;
1277 	char arg[20];
1278 	int ret, i;
1279 
1280 	if (cmdline_find_option_bool(boot_command_line, "nospectre_v2") ||
1281 	    cpu_mitigations_off())
1282 		return SPECTRE_V2_CMD_NONE;
1283 
1284 	ret = cmdline_find_option(boot_command_line, "spectre_v2", arg, sizeof(arg));
1285 	if (ret < 0)
1286 		return SPECTRE_V2_CMD_AUTO;
1287 
1288 	for (i = 0; i < ARRAY_SIZE(mitigation_options); i++) {
1289 		if (!match_option(arg, ret, mitigation_options[i].option))
1290 			continue;
1291 		cmd = mitigation_options[i].cmd;
1292 		break;
1293 	}
1294 
1295 	if (i >= ARRAY_SIZE(mitigation_options)) {
1296 		pr_err("unknown option (%s). Switching to AUTO select\n", arg);
1297 		return SPECTRE_V2_CMD_AUTO;
1298 	}
1299 
1300 	if ((cmd == SPECTRE_V2_CMD_RETPOLINE ||
1301 	     cmd == SPECTRE_V2_CMD_RETPOLINE_LFENCE ||
1302 	     cmd == SPECTRE_V2_CMD_RETPOLINE_GENERIC ||
1303 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE ||
1304 	     cmd == SPECTRE_V2_CMD_EIBRS_RETPOLINE) &&
1305 	    !IS_ENABLED(CONFIG_RETPOLINE)) {
1306 		pr_err("%s selected but not compiled in. Switching to AUTO select\n",
1307 		       mitigation_options[i].option);
1308 		return SPECTRE_V2_CMD_AUTO;
1309 	}
1310 
1311 	if ((cmd == SPECTRE_V2_CMD_EIBRS ||
1312 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE ||
1313 	     cmd == SPECTRE_V2_CMD_EIBRS_RETPOLINE) &&
1314 	    !boot_cpu_has(X86_FEATURE_IBRS_ENHANCED)) {
1315 		pr_err("%s selected but CPU doesn't have Enhanced or Automatic IBRS. Switching to AUTO select\n",
1316 		       mitigation_options[i].option);
1317 		return SPECTRE_V2_CMD_AUTO;
1318 	}
1319 
1320 	if ((cmd == SPECTRE_V2_CMD_RETPOLINE_LFENCE ||
1321 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE) &&
1322 	    !boot_cpu_has(X86_FEATURE_LFENCE_RDTSC)) {
1323 		pr_err("%s selected, but CPU doesn't have a serializing LFENCE. 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 && !IS_ENABLED(CONFIG_CPU_IBRS_ENTRY)) {
1329 		pr_err("%s selected but not compiled in. 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 && boot_cpu_data.x86_vendor != X86_VENDOR_INTEL) {
1335 		pr_err("%s selected but not Intel CPU. Switching to AUTO select\n",
1336 		       mitigation_options[i].option);
1337 		return SPECTRE_V2_CMD_AUTO;
1338 	}
1339 
1340 	if (cmd == SPECTRE_V2_CMD_IBRS && !boot_cpu_has(X86_FEATURE_IBRS)) {
1341 		pr_err("%s selected but CPU doesn't have IBRS. Switching to AUTO select\n",
1342 		       mitigation_options[i].option);
1343 		return SPECTRE_V2_CMD_AUTO;
1344 	}
1345 
1346 	if (cmd == SPECTRE_V2_CMD_IBRS && cpu_feature_enabled(X86_FEATURE_XENPV)) {
1347 		pr_err("%s selected but running as XenPV guest. Switching to AUTO select\n",
1348 		       mitigation_options[i].option);
1349 		return SPECTRE_V2_CMD_AUTO;
1350 	}
1351 
1352 	spec_v2_print_cond(mitigation_options[i].option,
1353 			   mitigation_options[i].secure);
1354 	return cmd;
1355 }
1356 
1357 static enum spectre_v2_mitigation __init spectre_v2_select_retpoline(void)
1358 {
1359 	if (!IS_ENABLED(CONFIG_RETPOLINE)) {
1360 		pr_err("Kernel not compiled with retpoline; no mitigation available!");
1361 		return SPECTRE_V2_NONE;
1362 	}
1363 
1364 	return SPECTRE_V2_RETPOLINE;
1365 }
1366 
1367 /* Disable in-kernel use of non-RSB RET predictors */
1368 static void __init spec_ctrl_disable_kernel_rrsba(void)
1369 {
1370 	u64 ia32_cap;
1371 
1372 	if (!boot_cpu_has(X86_FEATURE_RRSBA_CTRL))
1373 		return;
1374 
1375 	ia32_cap = x86_read_arch_cap_msr();
1376 
1377 	if (ia32_cap & ARCH_CAP_RRSBA) {
1378 		x86_spec_ctrl_base |= SPEC_CTRL_RRSBA_DIS_S;
1379 		update_spec_ctrl(x86_spec_ctrl_base);
1380 	}
1381 }
1382 
1383 static void __init spectre_v2_determine_rsb_fill_type_at_vmexit(enum spectre_v2_mitigation mode)
1384 {
1385 	/*
1386 	 * Similar to context switches, there are two types of RSB attacks
1387 	 * after VM exit:
1388 	 *
1389 	 * 1) RSB underflow
1390 	 *
1391 	 * 2) Poisoned RSB entry
1392 	 *
1393 	 * When retpoline is enabled, both are mitigated by filling/clearing
1394 	 * the RSB.
1395 	 *
1396 	 * When IBRS is enabled, while #1 would be mitigated by the IBRS branch
1397 	 * prediction isolation protections, RSB still needs to be cleared
1398 	 * because of #2.  Note that SMEP provides no protection here, unlike
1399 	 * user-space-poisoned RSB entries.
1400 	 *
1401 	 * eIBRS should protect against RSB poisoning, but if the EIBRS_PBRSB
1402 	 * bug is present then a LITE version of RSB protection is required,
1403 	 * just a single call needs to retire before a RET is executed.
1404 	 */
1405 	switch (mode) {
1406 	case SPECTRE_V2_NONE:
1407 		return;
1408 
1409 	case SPECTRE_V2_EIBRS_LFENCE:
1410 	case SPECTRE_V2_EIBRS:
1411 		if (boot_cpu_has_bug(X86_BUG_EIBRS_PBRSB)) {
1412 			setup_force_cpu_cap(X86_FEATURE_RSB_VMEXIT_LITE);
1413 			pr_info("Spectre v2 / PBRSB-eIBRS: Retire a single CALL on VMEXIT\n");
1414 		}
1415 		return;
1416 
1417 	case SPECTRE_V2_EIBRS_RETPOLINE:
1418 	case SPECTRE_V2_RETPOLINE:
1419 	case SPECTRE_V2_LFENCE:
1420 	case SPECTRE_V2_IBRS:
1421 		setup_force_cpu_cap(X86_FEATURE_RSB_VMEXIT);
1422 		pr_info("Spectre v2 / SpectreRSB : Filling RSB on VMEXIT\n");
1423 		return;
1424 	}
1425 
1426 	pr_warn_once("Unknown Spectre v2 mode, disabling RSB mitigation at VM exit");
1427 	dump_stack();
1428 }
1429 
1430 static void __init spectre_v2_select_mitigation(void)
1431 {
1432 	enum spectre_v2_mitigation_cmd cmd = spectre_v2_parse_cmdline();
1433 	enum spectre_v2_mitigation mode = SPECTRE_V2_NONE;
1434 
1435 	/*
1436 	 * If the CPU is not affected and the command line mode is NONE or AUTO
1437 	 * then nothing to do.
1438 	 */
1439 	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V2) &&
1440 	    (cmd == SPECTRE_V2_CMD_NONE || cmd == SPECTRE_V2_CMD_AUTO))
1441 		return;
1442 
1443 	switch (cmd) {
1444 	case SPECTRE_V2_CMD_NONE:
1445 		return;
1446 
1447 	case SPECTRE_V2_CMD_FORCE:
1448 	case SPECTRE_V2_CMD_AUTO:
1449 		if (boot_cpu_has(X86_FEATURE_IBRS_ENHANCED)) {
1450 			mode = SPECTRE_V2_EIBRS;
1451 			break;
1452 		}
1453 
1454 		if (IS_ENABLED(CONFIG_CPU_IBRS_ENTRY) &&
1455 		    boot_cpu_has_bug(X86_BUG_RETBLEED) &&
1456 		    retbleed_cmd != RETBLEED_CMD_OFF &&
1457 		    retbleed_cmd != RETBLEED_CMD_STUFF &&
1458 		    boot_cpu_has(X86_FEATURE_IBRS) &&
1459 		    boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) {
1460 			mode = SPECTRE_V2_IBRS;
1461 			break;
1462 		}
1463 
1464 		mode = spectre_v2_select_retpoline();
1465 		break;
1466 
1467 	case SPECTRE_V2_CMD_RETPOLINE_LFENCE:
1468 		pr_err(SPECTRE_V2_LFENCE_MSG);
1469 		mode = SPECTRE_V2_LFENCE;
1470 		break;
1471 
1472 	case SPECTRE_V2_CMD_RETPOLINE_GENERIC:
1473 		mode = SPECTRE_V2_RETPOLINE;
1474 		break;
1475 
1476 	case SPECTRE_V2_CMD_RETPOLINE:
1477 		mode = spectre_v2_select_retpoline();
1478 		break;
1479 
1480 	case SPECTRE_V2_CMD_IBRS:
1481 		mode = SPECTRE_V2_IBRS;
1482 		break;
1483 
1484 	case SPECTRE_V2_CMD_EIBRS:
1485 		mode = SPECTRE_V2_EIBRS;
1486 		break;
1487 
1488 	case SPECTRE_V2_CMD_EIBRS_LFENCE:
1489 		mode = SPECTRE_V2_EIBRS_LFENCE;
1490 		break;
1491 
1492 	case SPECTRE_V2_CMD_EIBRS_RETPOLINE:
1493 		mode = SPECTRE_V2_EIBRS_RETPOLINE;
1494 		break;
1495 	}
1496 
1497 	if (mode == SPECTRE_V2_EIBRS && unprivileged_ebpf_enabled())
1498 		pr_err(SPECTRE_V2_EIBRS_EBPF_MSG);
1499 
1500 	if (spectre_v2_in_ibrs_mode(mode)) {
1501 		if (boot_cpu_has(X86_FEATURE_AUTOIBRS)) {
1502 			msr_set_bit(MSR_EFER, _EFER_AUTOIBRS);
1503 		} else {
1504 			x86_spec_ctrl_base |= SPEC_CTRL_IBRS;
1505 			update_spec_ctrl(x86_spec_ctrl_base);
1506 		}
1507 	}
1508 
1509 	switch (mode) {
1510 	case SPECTRE_V2_NONE:
1511 	case SPECTRE_V2_EIBRS:
1512 		break;
1513 
1514 	case SPECTRE_V2_IBRS:
1515 		setup_force_cpu_cap(X86_FEATURE_KERNEL_IBRS);
1516 		if (boot_cpu_has(X86_FEATURE_IBRS_ENHANCED))
1517 			pr_warn(SPECTRE_V2_IBRS_PERF_MSG);
1518 		break;
1519 
1520 	case SPECTRE_V2_LFENCE:
1521 	case SPECTRE_V2_EIBRS_LFENCE:
1522 		setup_force_cpu_cap(X86_FEATURE_RETPOLINE_LFENCE);
1523 		fallthrough;
1524 
1525 	case SPECTRE_V2_RETPOLINE:
1526 	case SPECTRE_V2_EIBRS_RETPOLINE:
1527 		setup_force_cpu_cap(X86_FEATURE_RETPOLINE);
1528 		break;
1529 	}
1530 
1531 	/*
1532 	 * Disable alternate RSB predictions in kernel when indirect CALLs and
1533 	 * JMPs gets protection against BHI and Intramode-BTI, but RET
1534 	 * prediction from a non-RSB predictor is still a risk.
1535 	 */
1536 	if (mode == SPECTRE_V2_EIBRS_LFENCE ||
1537 	    mode == SPECTRE_V2_EIBRS_RETPOLINE ||
1538 	    mode == SPECTRE_V2_RETPOLINE)
1539 		spec_ctrl_disable_kernel_rrsba();
1540 
1541 	spectre_v2_enabled = mode;
1542 	pr_info("%s\n", spectre_v2_strings[mode]);
1543 
1544 	/*
1545 	 * If Spectre v2 protection has been enabled, fill the RSB during a
1546 	 * context switch.  In general there are two types of RSB attacks
1547 	 * across context switches, for which the CALLs/RETs may be unbalanced.
1548 	 *
1549 	 * 1) RSB underflow
1550 	 *
1551 	 *    Some Intel parts have "bottomless RSB".  When the RSB is empty,
1552 	 *    speculated return targets may come from the branch predictor,
1553 	 *    which could have a user-poisoned BTB or BHB entry.
1554 	 *
1555 	 *    AMD has it even worse: *all* returns are speculated from the BTB,
1556 	 *    regardless of the state of the RSB.
1557 	 *
1558 	 *    When IBRS or eIBRS is enabled, the "user -> kernel" attack
1559 	 *    scenario is mitigated by the IBRS branch prediction isolation
1560 	 *    properties, so the RSB buffer filling wouldn't be necessary to
1561 	 *    protect against this type of attack.
1562 	 *
1563 	 *    The "user -> user" attack scenario is mitigated by RSB filling.
1564 	 *
1565 	 * 2) Poisoned RSB entry
1566 	 *
1567 	 *    If the 'next' in-kernel return stack is shorter than 'prev',
1568 	 *    'next' could be tricked into speculating with a user-poisoned RSB
1569 	 *    entry.
1570 	 *
1571 	 *    The "user -> kernel" attack scenario is mitigated by SMEP and
1572 	 *    eIBRS.
1573 	 *
1574 	 *    The "user -> user" scenario, also known as SpectreBHB, requires
1575 	 *    RSB clearing.
1576 	 *
1577 	 * So to mitigate all cases, unconditionally fill RSB on context
1578 	 * switches.
1579 	 *
1580 	 * FIXME: Is this pointless for retbleed-affected AMD?
1581 	 */
1582 	setup_force_cpu_cap(X86_FEATURE_RSB_CTXSW);
1583 	pr_info("Spectre v2 / SpectreRSB mitigation: Filling RSB on context switch\n");
1584 
1585 	spectre_v2_determine_rsb_fill_type_at_vmexit(mode);
1586 
1587 	/*
1588 	 * Retpoline protects the kernel, but doesn't protect firmware.  IBRS
1589 	 * and Enhanced IBRS protect firmware too, so enable IBRS around
1590 	 * firmware calls only when IBRS / Enhanced / Automatic IBRS aren't
1591 	 * otherwise enabled.
1592 	 *
1593 	 * Use "mode" to check Enhanced IBRS instead of boot_cpu_has(), because
1594 	 * the user might select retpoline on the kernel command line and if
1595 	 * the CPU supports Enhanced IBRS, kernel might un-intentionally not
1596 	 * enable IBRS around firmware calls.
1597 	 */
1598 	if (boot_cpu_has_bug(X86_BUG_RETBLEED) &&
1599 	    boot_cpu_has(X86_FEATURE_IBPB) &&
1600 	    (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
1601 	     boot_cpu_data.x86_vendor == X86_VENDOR_HYGON)) {
1602 
1603 		if (retbleed_cmd != RETBLEED_CMD_IBPB) {
1604 			setup_force_cpu_cap(X86_FEATURE_USE_IBPB_FW);
1605 			pr_info("Enabling Speculation Barrier for firmware calls\n");
1606 		}
1607 
1608 	} else if (boot_cpu_has(X86_FEATURE_IBRS) && !spectre_v2_in_ibrs_mode(mode)) {
1609 		setup_force_cpu_cap(X86_FEATURE_USE_IBRS_FW);
1610 		pr_info("Enabling Restricted Speculation for firmware calls\n");
1611 	}
1612 
1613 	/* Set up IBPB and STIBP depending on the general spectre V2 command */
1614 	spectre_v2_cmd = cmd;
1615 }
1616 
1617 static void update_stibp_msr(void * __unused)
1618 {
1619 	u64 val = spec_ctrl_current() | (x86_spec_ctrl_base & SPEC_CTRL_STIBP);
1620 	update_spec_ctrl(val);
1621 }
1622 
1623 /* Update x86_spec_ctrl_base in case SMT state changed. */
1624 static void update_stibp_strict(void)
1625 {
1626 	u64 mask = x86_spec_ctrl_base & ~SPEC_CTRL_STIBP;
1627 
1628 	if (sched_smt_active())
1629 		mask |= SPEC_CTRL_STIBP;
1630 
1631 	if (mask == x86_spec_ctrl_base)
1632 		return;
1633 
1634 	pr_info("Update user space SMT mitigation: STIBP %s\n",
1635 		mask & SPEC_CTRL_STIBP ? "always-on" : "off");
1636 	x86_spec_ctrl_base = mask;
1637 	on_each_cpu(update_stibp_msr, NULL, 1);
1638 }
1639 
1640 /* Update the static key controlling the evaluation of TIF_SPEC_IB */
1641 static void update_indir_branch_cond(void)
1642 {
1643 	if (sched_smt_active())
1644 		static_branch_enable(&switch_to_cond_stibp);
1645 	else
1646 		static_branch_disable(&switch_to_cond_stibp);
1647 }
1648 
1649 #undef pr_fmt
1650 #define pr_fmt(fmt) fmt
1651 
1652 /* Update the static key controlling the MDS CPU buffer clear in idle */
1653 static void update_mds_branch_idle(void)
1654 {
1655 	u64 ia32_cap = x86_read_arch_cap_msr();
1656 
1657 	/*
1658 	 * Enable the idle clearing if SMT is active on CPUs which are
1659 	 * affected only by MSBDS and not any other MDS variant.
1660 	 *
1661 	 * The other variants cannot be mitigated when SMT is enabled, so
1662 	 * clearing the buffers on idle just to prevent the Store Buffer
1663 	 * repartitioning leak would be a window dressing exercise.
1664 	 */
1665 	if (!boot_cpu_has_bug(X86_BUG_MSBDS_ONLY))
1666 		return;
1667 
1668 	if (sched_smt_active()) {
1669 		static_branch_enable(&mds_idle_clear);
1670 	} else if (mmio_mitigation == MMIO_MITIGATION_OFF ||
1671 		   (ia32_cap & ARCH_CAP_FBSDP_NO)) {
1672 		static_branch_disable(&mds_idle_clear);
1673 	}
1674 }
1675 
1676 #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"
1677 #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"
1678 #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"
1679 
1680 void cpu_bugs_smt_update(void)
1681 {
1682 	mutex_lock(&spec_ctrl_mutex);
1683 
1684 	if (sched_smt_active() && unprivileged_ebpf_enabled() &&
1685 	    spectre_v2_enabled == SPECTRE_V2_EIBRS_LFENCE)
1686 		pr_warn_once(SPECTRE_V2_EIBRS_LFENCE_EBPF_SMT_MSG);
1687 
1688 	switch (spectre_v2_user_stibp) {
1689 	case SPECTRE_V2_USER_NONE:
1690 		break;
1691 	case SPECTRE_V2_USER_STRICT:
1692 	case SPECTRE_V2_USER_STRICT_PREFERRED:
1693 		update_stibp_strict();
1694 		break;
1695 	case SPECTRE_V2_USER_PRCTL:
1696 	case SPECTRE_V2_USER_SECCOMP:
1697 		update_indir_branch_cond();
1698 		break;
1699 	}
1700 
1701 	switch (mds_mitigation) {
1702 	case MDS_MITIGATION_FULL:
1703 	case MDS_MITIGATION_VMWERV:
1704 		if (sched_smt_active() && !boot_cpu_has(X86_BUG_MSBDS_ONLY))
1705 			pr_warn_once(MDS_MSG_SMT);
1706 		update_mds_branch_idle();
1707 		break;
1708 	case MDS_MITIGATION_OFF:
1709 		break;
1710 	}
1711 
1712 	switch (taa_mitigation) {
1713 	case TAA_MITIGATION_VERW:
1714 	case TAA_MITIGATION_UCODE_NEEDED:
1715 		if (sched_smt_active())
1716 			pr_warn_once(TAA_MSG_SMT);
1717 		break;
1718 	case TAA_MITIGATION_TSX_DISABLED:
1719 	case TAA_MITIGATION_OFF:
1720 		break;
1721 	}
1722 
1723 	switch (mmio_mitigation) {
1724 	case MMIO_MITIGATION_VERW:
1725 	case MMIO_MITIGATION_UCODE_NEEDED:
1726 		if (sched_smt_active())
1727 			pr_warn_once(MMIO_MSG_SMT);
1728 		break;
1729 	case MMIO_MITIGATION_OFF:
1730 		break;
1731 	}
1732 
1733 	mutex_unlock(&spec_ctrl_mutex);
1734 }
1735 
1736 #undef pr_fmt
1737 #define pr_fmt(fmt)	"Speculative Store Bypass: " fmt
1738 
1739 static enum ssb_mitigation ssb_mode __ro_after_init = SPEC_STORE_BYPASS_NONE;
1740 
1741 /* The kernel command line selection */
1742 enum ssb_mitigation_cmd {
1743 	SPEC_STORE_BYPASS_CMD_NONE,
1744 	SPEC_STORE_BYPASS_CMD_AUTO,
1745 	SPEC_STORE_BYPASS_CMD_ON,
1746 	SPEC_STORE_BYPASS_CMD_PRCTL,
1747 	SPEC_STORE_BYPASS_CMD_SECCOMP,
1748 };
1749 
1750 static const char * const ssb_strings[] = {
1751 	[SPEC_STORE_BYPASS_NONE]	= "Vulnerable",
1752 	[SPEC_STORE_BYPASS_DISABLE]	= "Mitigation: Speculative Store Bypass disabled",
1753 	[SPEC_STORE_BYPASS_PRCTL]	= "Mitigation: Speculative Store Bypass disabled via prctl",
1754 	[SPEC_STORE_BYPASS_SECCOMP]	= "Mitigation: Speculative Store Bypass disabled via prctl and seccomp",
1755 };
1756 
1757 static const struct {
1758 	const char *option;
1759 	enum ssb_mitigation_cmd cmd;
1760 } ssb_mitigation_options[]  __initconst = {
1761 	{ "auto",	SPEC_STORE_BYPASS_CMD_AUTO },    /* Platform decides */
1762 	{ "on",		SPEC_STORE_BYPASS_CMD_ON },      /* Disable Speculative Store Bypass */
1763 	{ "off",	SPEC_STORE_BYPASS_CMD_NONE },    /* Don't touch Speculative Store Bypass */
1764 	{ "prctl",	SPEC_STORE_BYPASS_CMD_PRCTL },   /* Disable Speculative Store Bypass via prctl */
1765 	{ "seccomp",	SPEC_STORE_BYPASS_CMD_SECCOMP }, /* Disable Speculative Store Bypass via prctl and seccomp */
1766 };
1767 
1768 static enum ssb_mitigation_cmd __init ssb_parse_cmdline(void)
1769 {
1770 	enum ssb_mitigation_cmd cmd = SPEC_STORE_BYPASS_CMD_AUTO;
1771 	char arg[20];
1772 	int ret, i;
1773 
1774 	if (cmdline_find_option_bool(boot_command_line, "nospec_store_bypass_disable") ||
1775 	    cpu_mitigations_off()) {
1776 		return SPEC_STORE_BYPASS_CMD_NONE;
1777 	} else {
1778 		ret = cmdline_find_option(boot_command_line, "spec_store_bypass_disable",
1779 					  arg, sizeof(arg));
1780 		if (ret < 0)
1781 			return SPEC_STORE_BYPASS_CMD_AUTO;
1782 
1783 		for (i = 0; i < ARRAY_SIZE(ssb_mitigation_options); i++) {
1784 			if (!match_option(arg, ret, ssb_mitigation_options[i].option))
1785 				continue;
1786 
1787 			cmd = ssb_mitigation_options[i].cmd;
1788 			break;
1789 		}
1790 
1791 		if (i >= ARRAY_SIZE(ssb_mitigation_options)) {
1792 			pr_err("unknown option (%s). Switching to AUTO select\n", arg);
1793 			return SPEC_STORE_BYPASS_CMD_AUTO;
1794 		}
1795 	}
1796 
1797 	return cmd;
1798 }
1799 
1800 static enum ssb_mitigation __init __ssb_select_mitigation(void)
1801 {
1802 	enum ssb_mitigation mode = SPEC_STORE_BYPASS_NONE;
1803 	enum ssb_mitigation_cmd cmd;
1804 
1805 	if (!boot_cpu_has(X86_FEATURE_SSBD))
1806 		return mode;
1807 
1808 	cmd = ssb_parse_cmdline();
1809 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS) &&
1810 	    (cmd == SPEC_STORE_BYPASS_CMD_NONE ||
1811 	     cmd == SPEC_STORE_BYPASS_CMD_AUTO))
1812 		return mode;
1813 
1814 	switch (cmd) {
1815 	case SPEC_STORE_BYPASS_CMD_SECCOMP:
1816 		/*
1817 		 * Choose prctl+seccomp as the default mode if seccomp is
1818 		 * enabled.
1819 		 */
1820 		if (IS_ENABLED(CONFIG_SECCOMP))
1821 			mode = SPEC_STORE_BYPASS_SECCOMP;
1822 		else
1823 			mode = SPEC_STORE_BYPASS_PRCTL;
1824 		break;
1825 	case SPEC_STORE_BYPASS_CMD_ON:
1826 		mode = SPEC_STORE_BYPASS_DISABLE;
1827 		break;
1828 	case SPEC_STORE_BYPASS_CMD_AUTO:
1829 	case SPEC_STORE_BYPASS_CMD_PRCTL:
1830 		mode = SPEC_STORE_BYPASS_PRCTL;
1831 		break;
1832 	case SPEC_STORE_BYPASS_CMD_NONE:
1833 		break;
1834 	}
1835 
1836 	/*
1837 	 * We have three CPU feature flags that are in play here:
1838 	 *  - X86_BUG_SPEC_STORE_BYPASS - CPU is susceptible.
1839 	 *  - X86_FEATURE_SSBD - CPU is able to turn off speculative store bypass
1840 	 *  - X86_FEATURE_SPEC_STORE_BYPASS_DISABLE - engage the mitigation
1841 	 */
1842 	if (mode == SPEC_STORE_BYPASS_DISABLE) {
1843 		setup_force_cpu_cap(X86_FEATURE_SPEC_STORE_BYPASS_DISABLE);
1844 		/*
1845 		 * Intel uses the SPEC CTRL MSR Bit(2) for this, while AMD may
1846 		 * use a completely different MSR and bit dependent on family.
1847 		 */
1848 		if (!static_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) &&
1849 		    !static_cpu_has(X86_FEATURE_AMD_SSBD)) {
1850 			x86_amd_ssb_disable();
1851 		} else {
1852 			x86_spec_ctrl_base |= SPEC_CTRL_SSBD;
1853 			update_spec_ctrl(x86_spec_ctrl_base);
1854 		}
1855 	}
1856 
1857 	return mode;
1858 }
1859 
1860 static void ssb_select_mitigation(void)
1861 {
1862 	ssb_mode = __ssb_select_mitigation();
1863 
1864 	if (boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1865 		pr_info("%s\n", ssb_strings[ssb_mode]);
1866 }
1867 
1868 #undef pr_fmt
1869 #define pr_fmt(fmt)     "Speculation prctl: " fmt
1870 
1871 static void task_update_spec_tif(struct task_struct *tsk)
1872 {
1873 	/* Force the update of the real TIF bits */
1874 	set_tsk_thread_flag(tsk, TIF_SPEC_FORCE_UPDATE);
1875 
1876 	/*
1877 	 * Immediately update the speculation control MSRs for the current
1878 	 * task, but for a non-current task delay setting the CPU
1879 	 * mitigation until it is scheduled next.
1880 	 *
1881 	 * This can only happen for SECCOMP mitigation. For PRCTL it's
1882 	 * always the current task.
1883 	 */
1884 	if (tsk == current)
1885 		speculation_ctrl_update_current();
1886 }
1887 
1888 static int l1d_flush_prctl_set(struct task_struct *task, unsigned long ctrl)
1889 {
1890 
1891 	if (!static_branch_unlikely(&switch_mm_cond_l1d_flush))
1892 		return -EPERM;
1893 
1894 	switch (ctrl) {
1895 	case PR_SPEC_ENABLE:
1896 		set_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH);
1897 		return 0;
1898 	case PR_SPEC_DISABLE:
1899 		clear_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH);
1900 		return 0;
1901 	default:
1902 		return -ERANGE;
1903 	}
1904 }
1905 
1906 static int ssb_prctl_set(struct task_struct *task, unsigned long ctrl)
1907 {
1908 	if (ssb_mode != SPEC_STORE_BYPASS_PRCTL &&
1909 	    ssb_mode != SPEC_STORE_BYPASS_SECCOMP)
1910 		return -ENXIO;
1911 
1912 	switch (ctrl) {
1913 	case PR_SPEC_ENABLE:
1914 		/* If speculation is force disabled, enable is not allowed */
1915 		if (task_spec_ssb_force_disable(task))
1916 			return -EPERM;
1917 		task_clear_spec_ssb_disable(task);
1918 		task_clear_spec_ssb_noexec(task);
1919 		task_update_spec_tif(task);
1920 		break;
1921 	case PR_SPEC_DISABLE:
1922 		task_set_spec_ssb_disable(task);
1923 		task_clear_spec_ssb_noexec(task);
1924 		task_update_spec_tif(task);
1925 		break;
1926 	case PR_SPEC_FORCE_DISABLE:
1927 		task_set_spec_ssb_disable(task);
1928 		task_set_spec_ssb_force_disable(task);
1929 		task_clear_spec_ssb_noexec(task);
1930 		task_update_spec_tif(task);
1931 		break;
1932 	case PR_SPEC_DISABLE_NOEXEC:
1933 		if (task_spec_ssb_force_disable(task))
1934 			return -EPERM;
1935 		task_set_spec_ssb_disable(task);
1936 		task_set_spec_ssb_noexec(task);
1937 		task_update_spec_tif(task);
1938 		break;
1939 	default:
1940 		return -ERANGE;
1941 	}
1942 	return 0;
1943 }
1944 
1945 static bool is_spec_ib_user_controlled(void)
1946 {
1947 	return spectre_v2_user_ibpb == SPECTRE_V2_USER_PRCTL ||
1948 		spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP ||
1949 		spectre_v2_user_stibp == SPECTRE_V2_USER_PRCTL ||
1950 		spectre_v2_user_stibp == SPECTRE_V2_USER_SECCOMP;
1951 }
1952 
1953 static int ib_prctl_set(struct task_struct *task, unsigned long ctrl)
1954 {
1955 	switch (ctrl) {
1956 	case PR_SPEC_ENABLE:
1957 		if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
1958 		    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
1959 			return 0;
1960 
1961 		/*
1962 		 * With strict mode for both IBPB and STIBP, the instruction
1963 		 * code paths avoid checking this task flag and instead,
1964 		 * unconditionally run the instruction. However, STIBP and IBPB
1965 		 * are independent and either can be set to conditionally
1966 		 * enabled regardless of the mode of the other.
1967 		 *
1968 		 * If either is set to conditional, allow the task flag to be
1969 		 * updated, unless it was force-disabled by a previous prctl
1970 		 * call. Currently, this is possible on an AMD CPU which has the
1971 		 * feature X86_FEATURE_AMD_STIBP_ALWAYS_ON. In this case, if the
1972 		 * kernel is booted with 'spectre_v2_user=seccomp', then
1973 		 * spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP and
1974 		 * spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED.
1975 		 */
1976 		if (!is_spec_ib_user_controlled() ||
1977 		    task_spec_ib_force_disable(task))
1978 			return -EPERM;
1979 
1980 		task_clear_spec_ib_disable(task);
1981 		task_update_spec_tif(task);
1982 		break;
1983 	case PR_SPEC_DISABLE:
1984 	case PR_SPEC_FORCE_DISABLE:
1985 		/*
1986 		 * Indirect branch speculation is always allowed when
1987 		 * mitigation is force disabled.
1988 		 */
1989 		if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
1990 		    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
1991 			return -EPERM;
1992 
1993 		if (!is_spec_ib_user_controlled())
1994 			return 0;
1995 
1996 		task_set_spec_ib_disable(task);
1997 		if (ctrl == PR_SPEC_FORCE_DISABLE)
1998 			task_set_spec_ib_force_disable(task);
1999 		task_update_spec_tif(task);
2000 		if (task == current)
2001 			indirect_branch_prediction_barrier();
2002 		break;
2003 	default:
2004 		return -ERANGE;
2005 	}
2006 	return 0;
2007 }
2008 
2009 int arch_prctl_spec_ctrl_set(struct task_struct *task, unsigned long which,
2010 			     unsigned long ctrl)
2011 {
2012 	switch (which) {
2013 	case PR_SPEC_STORE_BYPASS:
2014 		return ssb_prctl_set(task, ctrl);
2015 	case PR_SPEC_INDIRECT_BRANCH:
2016 		return ib_prctl_set(task, ctrl);
2017 	case PR_SPEC_L1D_FLUSH:
2018 		return l1d_flush_prctl_set(task, ctrl);
2019 	default:
2020 		return -ENODEV;
2021 	}
2022 }
2023 
2024 #ifdef CONFIG_SECCOMP
2025 void arch_seccomp_spec_mitigate(struct task_struct *task)
2026 {
2027 	if (ssb_mode == SPEC_STORE_BYPASS_SECCOMP)
2028 		ssb_prctl_set(task, PR_SPEC_FORCE_DISABLE);
2029 	if (spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP ||
2030 	    spectre_v2_user_stibp == SPECTRE_V2_USER_SECCOMP)
2031 		ib_prctl_set(task, PR_SPEC_FORCE_DISABLE);
2032 }
2033 #endif
2034 
2035 static int l1d_flush_prctl_get(struct task_struct *task)
2036 {
2037 	if (!static_branch_unlikely(&switch_mm_cond_l1d_flush))
2038 		return PR_SPEC_FORCE_DISABLE;
2039 
2040 	if (test_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH))
2041 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
2042 	else
2043 		return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
2044 }
2045 
2046 static int ssb_prctl_get(struct task_struct *task)
2047 {
2048 	switch (ssb_mode) {
2049 	case SPEC_STORE_BYPASS_DISABLE:
2050 		return PR_SPEC_DISABLE;
2051 	case SPEC_STORE_BYPASS_SECCOMP:
2052 	case SPEC_STORE_BYPASS_PRCTL:
2053 		if (task_spec_ssb_force_disable(task))
2054 			return PR_SPEC_PRCTL | PR_SPEC_FORCE_DISABLE;
2055 		if (task_spec_ssb_noexec(task))
2056 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE_NOEXEC;
2057 		if (task_spec_ssb_disable(task))
2058 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
2059 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
2060 	default:
2061 		if (boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
2062 			return PR_SPEC_ENABLE;
2063 		return PR_SPEC_NOT_AFFECTED;
2064 	}
2065 }
2066 
2067 static int ib_prctl_get(struct task_struct *task)
2068 {
2069 	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V2))
2070 		return PR_SPEC_NOT_AFFECTED;
2071 
2072 	if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
2073 	    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
2074 		return PR_SPEC_ENABLE;
2075 	else if (is_spec_ib_user_controlled()) {
2076 		if (task_spec_ib_force_disable(task))
2077 			return PR_SPEC_PRCTL | PR_SPEC_FORCE_DISABLE;
2078 		if (task_spec_ib_disable(task))
2079 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
2080 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
2081 	} else if (spectre_v2_user_ibpb == SPECTRE_V2_USER_STRICT ||
2082 	    spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT ||
2083 	    spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED)
2084 		return PR_SPEC_DISABLE;
2085 	else
2086 		return PR_SPEC_NOT_AFFECTED;
2087 }
2088 
2089 int arch_prctl_spec_ctrl_get(struct task_struct *task, unsigned long which)
2090 {
2091 	switch (which) {
2092 	case PR_SPEC_STORE_BYPASS:
2093 		return ssb_prctl_get(task);
2094 	case PR_SPEC_INDIRECT_BRANCH:
2095 		return ib_prctl_get(task);
2096 	case PR_SPEC_L1D_FLUSH:
2097 		return l1d_flush_prctl_get(task);
2098 	default:
2099 		return -ENODEV;
2100 	}
2101 }
2102 
2103 void x86_spec_ctrl_setup_ap(void)
2104 {
2105 	if (boot_cpu_has(X86_FEATURE_MSR_SPEC_CTRL))
2106 		update_spec_ctrl(x86_spec_ctrl_base);
2107 
2108 	if (ssb_mode == SPEC_STORE_BYPASS_DISABLE)
2109 		x86_amd_ssb_disable();
2110 }
2111 
2112 bool itlb_multihit_kvm_mitigation;
2113 EXPORT_SYMBOL_GPL(itlb_multihit_kvm_mitigation);
2114 
2115 #undef pr_fmt
2116 #define pr_fmt(fmt)	"L1TF: " fmt
2117 
2118 /* Default mitigation for L1TF-affected CPUs */
2119 enum l1tf_mitigations l1tf_mitigation __ro_after_init = L1TF_MITIGATION_FLUSH;
2120 #if IS_ENABLED(CONFIG_KVM_INTEL)
2121 EXPORT_SYMBOL_GPL(l1tf_mitigation);
2122 #endif
2123 enum vmx_l1d_flush_state l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
2124 EXPORT_SYMBOL_GPL(l1tf_vmx_mitigation);
2125 
2126 /*
2127  * These CPUs all support 44bits physical address space internally in the
2128  * cache but CPUID can report a smaller number of physical address bits.
2129  *
2130  * The L1TF mitigation uses the top most address bit for the inversion of
2131  * non present PTEs. When the installed memory reaches into the top most
2132  * address bit due to memory holes, which has been observed on machines
2133  * which report 36bits physical address bits and have 32G RAM installed,
2134  * then the mitigation range check in l1tf_select_mitigation() triggers.
2135  * This is a false positive because the mitigation is still possible due to
2136  * the fact that the cache uses 44bit internally. Use the cache bits
2137  * instead of the reported physical bits and adjust them on the affected
2138  * machines to 44bit if the reported bits are less than 44.
2139  */
2140 static void override_cache_bits(struct cpuinfo_x86 *c)
2141 {
2142 	if (c->x86 != 6)
2143 		return;
2144 
2145 	switch (c->x86_model) {
2146 	case INTEL_FAM6_NEHALEM:
2147 	case INTEL_FAM6_WESTMERE:
2148 	case INTEL_FAM6_SANDYBRIDGE:
2149 	case INTEL_FAM6_IVYBRIDGE:
2150 	case INTEL_FAM6_HASWELL:
2151 	case INTEL_FAM6_HASWELL_L:
2152 	case INTEL_FAM6_HASWELL_G:
2153 	case INTEL_FAM6_BROADWELL:
2154 	case INTEL_FAM6_BROADWELL_G:
2155 	case INTEL_FAM6_SKYLAKE_L:
2156 	case INTEL_FAM6_SKYLAKE:
2157 	case INTEL_FAM6_KABYLAKE_L:
2158 	case INTEL_FAM6_KABYLAKE:
2159 		if (c->x86_cache_bits < 44)
2160 			c->x86_cache_bits = 44;
2161 		break;
2162 	}
2163 }
2164 
2165 static void __init l1tf_select_mitigation(void)
2166 {
2167 	u64 half_pa;
2168 
2169 	if (!boot_cpu_has_bug(X86_BUG_L1TF))
2170 		return;
2171 
2172 	if (cpu_mitigations_off())
2173 		l1tf_mitigation = L1TF_MITIGATION_OFF;
2174 	else if (cpu_mitigations_auto_nosmt())
2175 		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOSMT;
2176 
2177 	override_cache_bits(&boot_cpu_data);
2178 
2179 	switch (l1tf_mitigation) {
2180 	case L1TF_MITIGATION_OFF:
2181 	case L1TF_MITIGATION_FLUSH_NOWARN:
2182 	case L1TF_MITIGATION_FLUSH:
2183 		break;
2184 	case L1TF_MITIGATION_FLUSH_NOSMT:
2185 	case L1TF_MITIGATION_FULL:
2186 		cpu_smt_disable(false);
2187 		break;
2188 	case L1TF_MITIGATION_FULL_FORCE:
2189 		cpu_smt_disable(true);
2190 		break;
2191 	}
2192 
2193 #if CONFIG_PGTABLE_LEVELS == 2
2194 	pr_warn("Kernel not compiled for PAE. No mitigation for L1TF\n");
2195 	return;
2196 #endif
2197 
2198 	half_pa = (u64)l1tf_pfn_limit() << PAGE_SHIFT;
2199 	if (l1tf_mitigation != L1TF_MITIGATION_OFF &&
2200 			e820__mapped_any(half_pa, ULLONG_MAX - half_pa, E820_TYPE_RAM)) {
2201 		pr_warn("System has more than MAX_PA/2 memory. L1TF mitigation not effective.\n");
2202 		pr_info("You may make it effective by booting the kernel with mem=%llu parameter.\n",
2203 				half_pa);
2204 		pr_info("However, doing so will make a part of your RAM unusable.\n");
2205 		pr_info("Reading https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/l1tf.html might help you decide.\n");
2206 		return;
2207 	}
2208 
2209 	setup_force_cpu_cap(X86_FEATURE_L1TF_PTEINV);
2210 }
2211 
2212 static int __init l1tf_cmdline(char *str)
2213 {
2214 	if (!boot_cpu_has_bug(X86_BUG_L1TF))
2215 		return 0;
2216 
2217 	if (!str)
2218 		return -EINVAL;
2219 
2220 	if (!strcmp(str, "off"))
2221 		l1tf_mitigation = L1TF_MITIGATION_OFF;
2222 	else if (!strcmp(str, "flush,nowarn"))
2223 		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOWARN;
2224 	else if (!strcmp(str, "flush"))
2225 		l1tf_mitigation = L1TF_MITIGATION_FLUSH;
2226 	else if (!strcmp(str, "flush,nosmt"))
2227 		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOSMT;
2228 	else if (!strcmp(str, "full"))
2229 		l1tf_mitigation = L1TF_MITIGATION_FULL;
2230 	else if (!strcmp(str, "full,force"))
2231 		l1tf_mitigation = L1TF_MITIGATION_FULL_FORCE;
2232 
2233 	return 0;
2234 }
2235 early_param("l1tf", l1tf_cmdline);
2236 
2237 #undef pr_fmt
2238 #define pr_fmt(fmt) fmt
2239 
2240 #ifdef CONFIG_SYSFS
2241 
2242 #define L1TF_DEFAULT_MSG "Mitigation: PTE Inversion"
2243 
2244 #if IS_ENABLED(CONFIG_KVM_INTEL)
2245 static const char * const l1tf_vmx_states[] = {
2246 	[VMENTER_L1D_FLUSH_AUTO]		= "auto",
2247 	[VMENTER_L1D_FLUSH_NEVER]		= "vulnerable",
2248 	[VMENTER_L1D_FLUSH_COND]		= "conditional cache flushes",
2249 	[VMENTER_L1D_FLUSH_ALWAYS]		= "cache flushes",
2250 	[VMENTER_L1D_FLUSH_EPT_DISABLED]	= "EPT disabled",
2251 	[VMENTER_L1D_FLUSH_NOT_REQUIRED]	= "flush not necessary"
2252 };
2253 
2254 static ssize_t l1tf_show_state(char *buf)
2255 {
2256 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO)
2257 		return sysfs_emit(buf, "%s\n", L1TF_DEFAULT_MSG);
2258 
2259 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_EPT_DISABLED ||
2260 	    (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER &&
2261 	     sched_smt_active())) {
2262 		return sysfs_emit(buf, "%s; VMX: %s\n", L1TF_DEFAULT_MSG,
2263 				  l1tf_vmx_states[l1tf_vmx_mitigation]);
2264 	}
2265 
2266 	return sysfs_emit(buf, "%s; VMX: %s, SMT %s\n", L1TF_DEFAULT_MSG,
2267 			  l1tf_vmx_states[l1tf_vmx_mitigation],
2268 			  sched_smt_active() ? "vulnerable" : "disabled");
2269 }
2270 
2271 static ssize_t itlb_multihit_show_state(char *buf)
2272 {
2273 	if (!boot_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
2274 	    !boot_cpu_has(X86_FEATURE_VMX))
2275 		return sysfs_emit(buf, "KVM: Mitigation: VMX unsupported\n");
2276 	else if (!(cr4_read_shadow() & X86_CR4_VMXE))
2277 		return sysfs_emit(buf, "KVM: Mitigation: VMX disabled\n");
2278 	else if (itlb_multihit_kvm_mitigation)
2279 		return sysfs_emit(buf, "KVM: Mitigation: Split huge pages\n");
2280 	else
2281 		return sysfs_emit(buf, "KVM: Vulnerable\n");
2282 }
2283 #else
2284 static ssize_t l1tf_show_state(char *buf)
2285 {
2286 	return sysfs_emit(buf, "%s\n", L1TF_DEFAULT_MSG);
2287 }
2288 
2289 static ssize_t itlb_multihit_show_state(char *buf)
2290 {
2291 	return sysfs_emit(buf, "Processor vulnerable\n");
2292 }
2293 #endif
2294 
2295 static ssize_t mds_show_state(char *buf)
2296 {
2297 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2298 		return sysfs_emit(buf, "%s; SMT Host state unknown\n",
2299 				  mds_strings[mds_mitigation]);
2300 	}
2301 
2302 	if (boot_cpu_has(X86_BUG_MSBDS_ONLY)) {
2303 		return sysfs_emit(buf, "%s; SMT %s\n", mds_strings[mds_mitigation],
2304 				  (mds_mitigation == MDS_MITIGATION_OFF ? "vulnerable" :
2305 				   sched_smt_active() ? "mitigated" : "disabled"));
2306 	}
2307 
2308 	return sysfs_emit(buf, "%s; SMT %s\n", mds_strings[mds_mitigation],
2309 			  sched_smt_active() ? "vulnerable" : "disabled");
2310 }
2311 
2312 static ssize_t tsx_async_abort_show_state(char *buf)
2313 {
2314 	if ((taa_mitigation == TAA_MITIGATION_TSX_DISABLED) ||
2315 	    (taa_mitigation == TAA_MITIGATION_OFF))
2316 		return sysfs_emit(buf, "%s\n", taa_strings[taa_mitigation]);
2317 
2318 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2319 		return sysfs_emit(buf, "%s; SMT Host state unknown\n",
2320 				  taa_strings[taa_mitigation]);
2321 	}
2322 
2323 	return sysfs_emit(buf, "%s; SMT %s\n", taa_strings[taa_mitigation],
2324 			  sched_smt_active() ? "vulnerable" : "disabled");
2325 }
2326 
2327 static ssize_t mmio_stale_data_show_state(char *buf)
2328 {
2329 	if (boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN))
2330 		return sysfs_emit(buf, "Unknown: No mitigations\n");
2331 
2332 	if (mmio_mitigation == MMIO_MITIGATION_OFF)
2333 		return sysfs_emit(buf, "%s\n", mmio_strings[mmio_mitigation]);
2334 
2335 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2336 		return sysfs_emit(buf, "%s; SMT Host state unknown\n",
2337 				  mmio_strings[mmio_mitigation]);
2338 	}
2339 
2340 	return sysfs_emit(buf, "%s; SMT %s\n", mmio_strings[mmio_mitigation],
2341 			  sched_smt_active() ? "vulnerable" : "disabled");
2342 }
2343 
2344 static char *stibp_state(void)
2345 {
2346 	if (spectre_v2_in_eibrs_mode(spectre_v2_enabled))
2347 		return "";
2348 
2349 	switch (spectre_v2_user_stibp) {
2350 	case SPECTRE_V2_USER_NONE:
2351 		return ", STIBP: disabled";
2352 	case SPECTRE_V2_USER_STRICT:
2353 		return ", STIBP: forced";
2354 	case SPECTRE_V2_USER_STRICT_PREFERRED:
2355 		return ", STIBP: always-on";
2356 	case SPECTRE_V2_USER_PRCTL:
2357 	case SPECTRE_V2_USER_SECCOMP:
2358 		if (static_key_enabled(&switch_to_cond_stibp))
2359 			return ", STIBP: conditional";
2360 	}
2361 	return "";
2362 }
2363 
2364 static char *ibpb_state(void)
2365 {
2366 	if (boot_cpu_has(X86_FEATURE_IBPB)) {
2367 		if (static_key_enabled(&switch_mm_always_ibpb))
2368 			return ", IBPB: always-on";
2369 		if (static_key_enabled(&switch_mm_cond_ibpb))
2370 			return ", IBPB: conditional";
2371 		return ", IBPB: disabled";
2372 	}
2373 	return "";
2374 }
2375 
2376 static char *pbrsb_eibrs_state(void)
2377 {
2378 	if (boot_cpu_has_bug(X86_BUG_EIBRS_PBRSB)) {
2379 		if (boot_cpu_has(X86_FEATURE_RSB_VMEXIT_LITE) ||
2380 		    boot_cpu_has(X86_FEATURE_RSB_VMEXIT))
2381 			return ", PBRSB-eIBRS: SW sequence";
2382 		else
2383 			return ", PBRSB-eIBRS: Vulnerable";
2384 	} else {
2385 		return ", PBRSB-eIBRS: Not affected";
2386 	}
2387 }
2388 
2389 static ssize_t spectre_v2_show_state(char *buf)
2390 {
2391 	if (spectre_v2_enabled == SPECTRE_V2_LFENCE)
2392 		return sysfs_emit(buf, "Vulnerable: LFENCE\n");
2393 
2394 	if (spectre_v2_enabled == SPECTRE_V2_EIBRS && unprivileged_ebpf_enabled())
2395 		return sysfs_emit(buf, "Vulnerable: eIBRS with unprivileged eBPF\n");
2396 
2397 	if (sched_smt_active() && unprivileged_ebpf_enabled() &&
2398 	    spectre_v2_enabled == SPECTRE_V2_EIBRS_LFENCE)
2399 		return sysfs_emit(buf, "Vulnerable: eIBRS+LFENCE with unprivileged eBPF and SMT\n");
2400 
2401 	return sysfs_emit(buf, "%s%s%s%s%s%s%s\n",
2402 			  spectre_v2_strings[spectre_v2_enabled],
2403 			  ibpb_state(),
2404 			  boot_cpu_has(X86_FEATURE_USE_IBRS_FW) ? ", IBRS_FW" : "",
2405 			  stibp_state(),
2406 			  boot_cpu_has(X86_FEATURE_RSB_CTXSW) ? ", RSB filling" : "",
2407 			  pbrsb_eibrs_state(),
2408 			  spectre_v2_module_string());
2409 }
2410 
2411 static ssize_t srbds_show_state(char *buf)
2412 {
2413 	return sysfs_emit(buf, "%s\n", srbds_strings[srbds_mitigation]);
2414 }
2415 
2416 static ssize_t retbleed_show_state(char *buf)
2417 {
2418 	if (retbleed_mitigation == RETBLEED_MITIGATION_UNRET ||
2419 	    retbleed_mitigation == RETBLEED_MITIGATION_IBPB) {
2420 		if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD &&
2421 		    boot_cpu_data.x86_vendor != X86_VENDOR_HYGON)
2422 			return sysfs_emit(buf, "Vulnerable: untrained return thunk / IBPB on non-AMD based uarch\n");
2423 
2424 		return sysfs_emit(buf, "%s; SMT %s\n", retbleed_strings[retbleed_mitigation],
2425 				  !sched_smt_active() ? "disabled" :
2426 				  spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT ||
2427 				  spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED ?
2428 				  "enabled with STIBP protection" : "vulnerable");
2429 	}
2430 
2431 	return sysfs_emit(buf, "%s\n", retbleed_strings[retbleed_mitigation]);
2432 }
2433 
2434 static ssize_t cpu_show_common(struct device *dev, struct device_attribute *attr,
2435 			       char *buf, unsigned int bug)
2436 {
2437 	if (!boot_cpu_has_bug(bug))
2438 		return sysfs_emit(buf, "Not affected\n");
2439 
2440 	switch (bug) {
2441 	case X86_BUG_CPU_MELTDOWN:
2442 		if (boot_cpu_has(X86_FEATURE_PTI))
2443 			return sysfs_emit(buf, "Mitigation: PTI\n");
2444 
2445 		if (hypervisor_is_type(X86_HYPER_XEN_PV))
2446 			return sysfs_emit(buf, "Unknown (XEN PV detected, hypervisor mitigation required)\n");
2447 
2448 		break;
2449 
2450 	case X86_BUG_SPECTRE_V1:
2451 		return sysfs_emit(buf, "%s\n", spectre_v1_strings[spectre_v1_mitigation]);
2452 
2453 	case X86_BUG_SPECTRE_V2:
2454 		return spectre_v2_show_state(buf);
2455 
2456 	case X86_BUG_SPEC_STORE_BYPASS:
2457 		return sysfs_emit(buf, "%s\n", ssb_strings[ssb_mode]);
2458 
2459 	case X86_BUG_L1TF:
2460 		if (boot_cpu_has(X86_FEATURE_L1TF_PTEINV))
2461 			return l1tf_show_state(buf);
2462 		break;
2463 
2464 	case X86_BUG_MDS:
2465 		return mds_show_state(buf);
2466 
2467 	case X86_BUG_TAA:
2468 		return tsx_async_abort_show_state(buf);
2469 
2470 	case X86_BUG_ITLB_MULTIHIT:
2471 		return itlb_multihit_show_state(buf);
2472 
2473 	case X86_BUG_SRBDS:
2474 		return srbds_show_state(buf);
2475 
2476 	case X86_BUG_MMIO_STALE_DATA:
2477 	case X86_BUG_MMIO_UNKNOWN:
2478 		return mmio_stale_data_show_state(buf);
2479 
2480 	case X86_BUG_RETBLEED:
2481 		return retbleed_show_state(buf);
2482 
2483 	default:
2484 		break;
2485 	}
2486 
2487 	return sysfs_emit(buf, "Vulnerable\n");
2488 }
2489 
2490 ssize_t cpu_show_meltdown(struct device *dev, struct device_attribute *attr, char *buf)
2491 {
2492 	return cpu_show_common(dev, attr, buf, X86_BUG_CPU_MELTDOWN);
2493 }
2494 
2495 ssize_t cpu_show_spectre_v1(struct device *dev, struct device_attribute *attr, char *buf)
2496 {
2497 	return cpu_show_common(dev, attr, buf, X86_BUG_SPECTRE_V1);
2498 }
2499 
2500 ssize_t cpu_show_spectre_v2(struct device *dev, struct device_attribute *attr, char *buf)
2501 {
2502 	return cpu_show_common(dev, attr, buf, X86_BUG_SPECTRE_V2);
2503 }
2504 
2505 ssize_t cpu_show_spec_store_bypass(struct device *dev, struct device_attribute *attr, char *buf)
2506 {
2507 	return cpu_show_common(dev, attr, buf, X86_BUG_SPEC_STORE_BYPASS);
2508 }
2509 
2510 ssize_t cpu_show_l1tf(struct device *dev, struct device_attribute *attr, char *buf)
2511 {
2512 	return cpu_show_common(dev, attr, buf, X86_BUG_L1TF);
2513 }
2514 
2515 ssize_t cpu_show_mds(struct device *dev, struct device_attribute *attr, char *buf)
2516 {
2517 	return cpu_show_common(dev, attr, buf, X86_BUG_MDS);
2518 }
2519 
2520 ssize_t cpu_show_tsx_async_abort(struct device *dev, struct device_attribute *attr, char *buf)
2521 {
2522 	return cpu_show_common(dev, attr, buf, X86_BUG_TAA);
2523 }
2524 
2525 ssize_t cpu_show_itlb_multihit(struct device *dev, struct device_attribute *attr, char *buf)
2526 {
2527 	return cpu_show_common(dev, attr, buf, X86_BUG_ITLB_MULTIHIT);
2528 }
2529 
2530 ssize_t cpu_show_srbds(struct device *dev, struct device_attribute *attr, char *buf)
2531 {
2532 	return cpu_show_common(dev, attr, buf, X86_BUG_SRBDS);
2533 }
2534 
2535 ssize_t cpu_show_mmio_stale_data(struct device *dev, struct device_attribute *attr, char *buf)
2536 {
2537 	if (boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN))
2538 		return cpu_show_common(dev, attr, buf, X86_BUG_MMIO_UNKNOWN);
2539 	else
2540 		return cpu_show_common(dev, attr, buf, X86_BUG_MMIO_STALE_DATA);
2541 }
2542 
2543 ssize_t cpu_show_retbleed(struct device *dev, struct device_attribute *attr, char *buf)
2544 {
2545 	return cpu_show_common(dev, attr, buf, X86_BUG_RETBLEED);
2546 }
2547 #endif
2548