• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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/internal.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 
52 /* The base value of the SPEC_CTRL MSR without task-specific bits set */
53 u64 x86_spec_ctrl_base;
54 EXPORT_SYMBOL_GPL(x86_spec_ctrl_base);
55 
56 /* The current value of the SPEC_CTRL MSR with task-specific bits set */
57 DEFINE_PER_CPU(u64, x86_spec_ctrl_current);
58 EXPORT_SYMBOL_GPL(x86_spec_ctrl_current);
59 
60 static DEFINE_MUTEX(spec_ctrl_mutex);
61 
62 /* Update SPEC_CTRL MSR and its cached copy unconditionally */
update_spec_ctrl(u64 val)63 static void update_spec_ctrl(u64 val)
64 {
65 	this_cpu_write(x86_spec_ctrl_current, val);
66 	wrmsrl(MSR_IA32_SPEC_CTRL, val);
67 }
68 
69 /*
70  * Keep track of the SPEC_CTRL MSR value for the current task, which may differ
71  * from x86_spec_ctrl_base due to STIBP/SSB in __speculation_ctrl_update().
72  */
update_spec_ctrl_cond(u64 val)73 void update_spec_ctrl_cond(u64 val)
74 {
75 	if (this_cpu_read(x86_spec_ctrl_current) == val)
76 		return;
77 
78 	this_cpu_write(x86_spec_ctrl_current, val);
79 
80 	/*
81 	 * When KERNEL_IBRS this MSR is written on return-to-user, unless
82 	 * forced the update can be delayed until that time.
83 	 */
84 	if (!cpu_feature_enabled(X86_FEATURE_KERNEL_IBRS))
85 		wrmsrl(MSR_IA32_SPEC_CTRL, val);
86 }
87 
spec_ctrl_current(void)88 u64 spec_ctrl_current(void)
89 {
90 	return this_cpu_read(x86_spec_ctrl_current);
91 }
92 EXPORT_SYMBOL_GPL(spec_ctrl_current);
93 
94 /*
95  * AMD specific MSR info for Speculative Store Bypass control.
96  * x86_amd_ls_cfg_ssbd_mask is initialized in identify_boot_cpu().
97  */
98 u64 __ro_after_init x86_amd_ls_cfg_base;
99 u64 __ro_after_init x86_amd_ls_cfg_ssbd_mask;
100 
101 /* Control conditional STIBP in switch_to() */
102 DEFINE_STATIC_KEY_FALSE(switch_to_cond_stibp);
103 /* Control conditional IBPB in switch_mm() */
104 DEFINE_STATIC_KEY_FALSE(switch_mm_cond_ibpb);
105 /* Control unconditional IBPB in switch_mm() */
106 DEFINE_STATIC_KEY_FALSE(switch_mm_always_ibpb);
107 
108 /* Control MDS CPU buffer clear before returning to user space */
109 DEFINE_STATIC_KEY_FALSE(mds_user_clear);
110 EXPORT_SYMBOL_GPL(mds_user_clear);
111 /* Control MDS CPU buffer clear before idling (halt, mwait) */
112 DEFINE_STATIC_KEY_FALSE(mds_idle_clear);
113 EXPORT_SYMBOL_GPL(mds_idle_clear);
114 
115 /* Controls CPU Fill buffer clear before KVM guest MMIO accesses */
116 DEFINE_STATIC_KEY_FALSE(mmio_stale_data_clear);
117 EXPORT_SYMBOL_GPL(mmio_stale_data_clear);
118 
check_bugs(void)119 void __init check_bugs(void)
120 {
121 	identify_boot_cpu();
122 
123 	/*
124 	 * identify_boot_cpu() initialized SMT support information, let the
125 	 * core code know.
126 	 */
127 	cpu_smt_check_topology();
128 
129 	if (!IS_ENABLED(CONFIG_SMP)) {
130 		pr_info("CPU: ");
131 		print_cpu_info(&boot_cpu_data);
132 	}
133 
134 	/*
135 	 * Read the SPEC_CTRL MSR to account for reserved bits which may
136 	 * have unknown values. AMD64_LS_CFG MSR is cached in the early AMD
137 	 * init code as it is not enumerated and depends on the family.
138 	 */
139 	if (boot_cpu_has(X86_FEATURE_MSR_SPEC_CTRL))
140 		rdmsrl(MSR_IA32_SPEC_CTRL, x86_spec_ctrl_base);
141 
142 	/* Select the proper CPU mitigations before patching alternatives: */
143 	spectre_v1_select_mitigation();
144 	spectre_v2_select_mitigation();
145 	/*
146 	 * retbleed_select_mitigation() relies on the state set by
147 	 * spectre_v2_select_mitigation(); specifically it wants to know about
148 	 * spectre_v2=ibrs.
149 	 */
150 	retbleed_select_mitigation();
151 	/*
152 	 * spectre_v2_user_select_mitigation() relies on the state set by
153 	 * retbleed_select_mitigation(); specifically the STIBP selection is
154 	 * forced for UNRET or IBPB.
155 	 */
156 	spectre_v2_user_select_mitigation();
157 	ssb_select_mitigation();
158 	l1tf_select_mitigation();
159 	md_clear_select_mitigation();
160 	srbds_select_mitigation();
161 
162 	arch_smt_update();
163 
164 #ifdef CONFIG_X86_32
165 	/*
166 	 * Check whether we are able to run this kernel safely on SMP.
167 	 *
168 	 * - i386 is no longer supported.
169 	 * - In order to run on anything without a TSC, we need to be
170 	 *   compiled for a i486.
171 	 */
172 	if (boot_cpu_data.x86 < 4)
173 		panic("Kernel requires i486+ for 'invlpg' and other features");
174 
175 	init_utsname()->machine[1] =
176 		'0' + (boot_cpu_data.x86 > 6 ? 6 : boot_cpu_data.x86);
177 	alternative_instructions();
178 
179 	fpu__init_check_bugs();
180 #else /* CONFIG_X86_64 */
181 	alternative_instructions();
182 
183 	/*
184 	 * Make sure the first 2MB area is not mapped by huge pages
185 	 * There are typically fixed size MTRRs in there and overlapping
186 	 * MTRRs into large pages causes slow downs.
187 	 *
188 	 * Right now we don't do that with gbpages because there seems
189 	 * very little benefit for that case.
190 	 */
191 	if (!direct_gbpages)
192 		set_memory_4k((unsigned long)__va(0), 1);
193 #endif
194 }
195 
196 /*
197  * NOTE: For VMX, this function is not called in the vmexit path.
198  * It uses vmx_spec_ctrl_restore_host() instead.
199  */
200 void
x86_virt_spec_ctrl(u64 guest_spec_ctrl,u64 guest_virt_spec_ctrl,bool setguest)201 x86_virt_spec_ctrl(u64 guest_spec_ctrl, u64 guest_virt_spec_ctrl, bool setguest)
202 {
203 	u64 msrval, guestval = guest_spec_ctrl, hostval = spec_ctrl_current();
204 	struct thread_info *ti = current_thread_info();
205 
206 	if (static_cpu_has(X86_FEATURE_MSR_SPEC_CTRL)) {
207 		if (hostval != guestval) {
208 			msrval = setguest ? guestval : hostval;
209 			wrmsrl(MSR_IA32_SPEC_CTRL, msrval);
210 		}
211 	}
212 
213 	/*
214 	 * If SSBD is not handled in MSR_SPEC_CTRL on AMD, update
215 	 * MSR_AMD64_L2_CFG or MSR_VIRT_SPEC_CTRL if supported.
216 	 */
217 	if (!static_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
218 	    !static_cpu_has(X86_FEATURE_VIRT_SSBD))
219 		return;
220 
221 	/*
222 	 * If the host has SSBD mitigation enabled, force it in the host's
223 	 * virtual MSR value. If its not permanently enabled, evaluate
224 	 * current's TIF_SSBD thread flag.
225 	 */
226 	if (static_cpu_has(X86_FEATURE_SPEC_STORE_BYPASS_DISABLE))
227 		hostval = SPEC_CTRL_SSBD;
228 	else
229 		hostval = ssbd_tif_to_spec_ctrl(ti->flags);
230 
231 	/* Sanitize the guest value */
232 	guestval = guest_virt_spec_ctrl & SPEC_CTRL_SSBD;
233 
234 	if (hostval != guestval) {
235 		unsigned long tif;
236 
237 		tif = setguest ? ssbd_spec_ctrl_to_tif(guestval) :
238 				 ssbd_spec_ctrl_to_tif(hostval);
239 
240 		speculation_ctrl_update(tif);
241 	}
242 }
243 EXPORT_SYMBOL_GPL(x86_virt_spec_ctrl);
244 
x86_amd_ssb_disable(void)245 static void x86_amd_ssb_disable(void)
246 {
247 	u64 msrval = x86_amd_ls_cfg_base | x86_amd_ls_cfg_ssbd_mask;
248 
249 	if (boot_cpu_has(X86_FEATURE_VIRT_SSBD))
250 		wrmsrl(MSR_AMD64_VIRT_SPEC_CTRL, SPEC_CTRL_SSBD);
251 	else if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD))
252 		wrmsrl(MSR_AMD64_LS_CFG, msrval);
253 }
254 
255 #undef pr_fmt
256 #define pr_fmt(fmt)	"MDS: " fmt
257 
258 /* Default mitigation for MDS-affected CPUs */
259 static enum mds_mitigations mds_mitigation __ro_after_init = MDS_MITIGATION_FULL;
260 static bool mds_nosmt __ro_after_init = false;
261 
262 static const char * const mds_strings[] = {
263 	[MDS_MITIGATION_OFF]	= "Vulnerable",
264 	[MDS_MITIGATION_FULL]	= "Mitigation: Clear CPU buffers",
265 	[MDS_MITIGATION_VMWERV]	= "Vulnerable: Clear CPU buffers attempted, no microcode",
266 };
267 
mds_select_mitigation(void)268 static void __init mds_select_mitigation(void)
269 {
270 	if (!boot_cpu_has_bug(X86_BUG_MDS) || cpu_mitigations_off()) {
271 		mds_mitigation = MDS_MITIGATION_OFF;
272 		return;
273 	}
274 
275 	if (mds_mitigation == MDS_MITIGATION_FULL) {
276 		if (!boot_cpu_has(X86_FEATURE_MD_CLEAR))
277 			mds_mitigation = MDS_MITIGATION_VMWERV;
278 
279 		static_branch_enable(&mds_user_clear);
280 
281 		if (!boot_cpu_has(X86_BUG_MSBDS_ONLY) &&
282 		    (mds_nosmt || cpu_mitigations_auto_nosmt()))
283 			cpu_smt_disable(false);
284 	}
285 }
286 
mds_cmdline(char * str)287 static int __init mds_cmdline(char *str)
288 {
289 	if (!boot_cpu_has_bug(X86_BUG_MDS))
290 		return 0;
291 
292 	if (!str)
293 		return -EINVAL;
294 
295 	if (!strcmp(str, "off"))
296 		mds_mitigation = MDS_MITIGATION_OFF;
297 	else if (!strcmp(str, "full"))
298 		mds_mitigation = MDS_MITIGATION_FULL;
299 	else if (!strcmp(str, "full,nosmt")) {
300 		mds_mitigation = MDS_MITIGATION_FULL;
301 		mds_nosmt = true;
302 	}
303 
304 	return 0;
305 }
306 early_param("mds", mds_cmdline);
307 
308 #undef pr_fmt
309 #define pr_fmt(fmt)	"TAA: " fmt
310 
311 enum taa_mitigations {
312 	TAA_MITIGATION_OFF,
313 	TAA_MITIGATION_UCODE_NEEDED,
314 	TAA_MITIGATION_VERW,
315 	TAA_MITIGATION_TSX_DISABLED,
316 };
317 
318 /* Default mitigation for TAA-affected CPUs */
319 static enum taa_mitigations taa_mitigation __ro_after_init = TAA_MITIGATION_VERW;
320 static bool taa_nosmt __ro_after_init;
321 
322 static const char * const taa_strings[] = {
323 	[TAA_MITIGATION_OFF]		= "Vulnerable",
324 	[TAA_MITIGATION_UCODE_NEEDED]	= "Vulnerable: Clear CPU buffers attempted, no microcode",
325 	[TAA_MITIGATION_VERW]		= "Mitigation: Clear CPU buffers",
326 	[TAA_MITIGATION_TSX_DISABLED]	= "Mitigation: TSX disabled",
327 };
328 
taa_select_mitigation(void)329 static void __init taa_select_mitigation(void)
330 {
331 	u64 ia32_cap;
332 
333 	if (!boot_cpu_has_bug(X86_BUG_TAA)) {
334 		taa_mitigation = TAA_MITIGATION_OFF;
335 		return;
336 	}
337 
338 	/* TSX previously disabled by tsx=off */
339 	if (!boot_cpu_has(X86_FEATURE_RTM)) {
340 		taa_mitigation = TAA_MITIGATION_TSX_DISABLED;
341 		return;
342 	}
343 
344 	if (cpu_mitigations_off()) {
345 		taa_mitigation = TAA_MITIGATION_OFF;
346 		return;
347 	}
348 
349 	/*
350 	 * TAA mitigation via VERW is turned off if both
351 	 * tsx_async_abort=off and mds=off are specified.
352 	 */
353 	if (taa_mitigation == TAA_MITIGATION_OFF &&
354 	    mds_mitigation == MDS_MITIGATION_OFF)
355 		return;
356 
357 	if (boot_cpu_has(X86_FEATURE_MD_CLEAR))
358 		taa_mitigation = TAA_MITIGATION_VERW;
359 	else
360 		taa_mitigation = TAA_MITIGATION_UCODE_NEEDED;
361 
362 	/*
363 	 * VERW doesn't clear the CPU buffers when MD_CLEAR=1 and MDS_NO=1.
364 	 * A microcode update fixes this behavior to clear CPU buffers. It also
365 	 * adds support for MSR_IA32_TSX_CTRL which is enumerated by the
366 	 * ARCH_CAP_TSX_CTRL_MSR bit.
367 	 *
368 	 * On MDS_NO=1 CPUs if ARCH_CAP_TSX_CTRL_MSR is not set, microcode
369 	 * update is required.
370 	 */
371 	ia32_cap = x86_read_arch_cap_msr();
372 	if ( (ia32_cap & ARCH_CAP_MDS_NO) &&
373 	    !(ia32_cap & ARCH_CAP_TSX_CTRL_MSR))
374 		taa_mitigation = TAA_MITIGATION_UCODE_NEEDED;
375 
376 	/*
377 	 * TSX is enabled, select alternate mitigation for TAA which is
378 	 * the same as MDS. Enable MDS static branch to clear CPU buffers.
379 	 *
380 	 * For guests that can't determine whether the correct microcode is
381 	 * present on host, enable the mitigation for UCODE_NEEDED as well.
382 	 */
383 	static_branch_enable(&mds_user_clear);
384 
385 	if (taa_nosmt || cpu_mitigations_auto_nosmt())
386 		cpu_smt_disable(false);
387 }
388 
tsx_async_abort_parse_cmdline(char * str)389 static int __init tsx_async_abort_parse_cmdline(char *str)
390 {
391 	if (!boot_cpu_has_bug(X86_BUG_TAA))
392 		return 0;
393 
394 	if (!str)
395 		return -EINVAL;
396 
397 	if (!strcmp(str, "off")) {
398 		taa_mitigation = TAA_MITIGATION_OFF;
399 	} else if (!strcmp(str, "full")) {
400 		taa_mitigation = TAA_MITIGATION_VERW;
401 	} else if (!strcmp(str, "full,nosmt")) {
402 		taa_mitigation = TAA_MITIGATION_VERW;
403 		taa_nosmt = true;
404 	}
405 
406 	return 0;
407 }
408 early_param("tsx_async_abort", tsx_async_abort_parse_cmdline);
409 
410 #undef pr_fmt
411 #define pr_fmt(fmt)	"MMIO Stale Data: " fmt
412 
413 enum mmio_mitigations {
414 	MMIO_MITIGATION_OFF,
415 	MMIO_MITIGATION_UCODE_NEEDED,
416 	MMIO_MITIGATION_VERW,
417 };
418 
419 /* Default mitigation for Processor MMIO Stale Data vulnerabilities */
420 static enum mmio_mitigations mmio_mitigation __ro_after_init = MMIO_MITIGATION_VERW;
421 static bool mmio_nosmt __ro_after_init = false;
422 
423 static const char * const mmio_strings[] = {
424 	[MMIO_MITIGATION_OFF]		= "Vulnerable",
425 	[MMIO_MITIGATION_UCODE_NEEDED]	= "Vulnerable: Clear CPU buffers attempted, no microcode",
426 	[MMIO_MITIGATION_VERW]		= "Mitigation: Clear CPU buffers",
427 };
428 
mmio_select_mitigation(void)429 static void __init mmio_select_mitigation(void)
430 {
431 	u64 ia32_cap;
432 
433 	if (!boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA) ||
434 	     boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN) ||
435 	     cpu_mitigations_off()) {
436 		mmio_mitigation = MMIO_MITIGATION_OFF;
437 		return;
438 	}
439 
440 	if (mmio_mitigation == MMIO_MITIGATION_OFF)
441 		return;
442 
443 	ia32_cap = x86_read_arch_cap_msr();
444 
445 	/*
446 	 * Enable CPU buffer clear mitigation for host and VMM, if also affected
447 	 * by MDS or TAA. Otherwise, enable mitigation for VMM only.
448 	 */
449 	if (boot_cpu_has_bug(X86_BUG_MDS) || (boot_cpu_has_bug(X86_BUG_TAA) &&
450 					      boot_cpu_has(X86_FEATURE_RTM)))
451 		static_branch_enable(&mds_user_clear);
452 	else
453 		static_branch_enable(&mmio_stale_data_clear);
454 
455 	/*
456 	 * If Processor-MMIO-Stale-Data bug is present and Fill Buffer data can
457 	 * be propagated to uncore buffers, clearing the Fill buffers on idle
458 	 * is required irrespective of SMT state.
459 	 */
460 	if (!(ia32_cap & ARCH_CAP_FBSDP_NO))
461 		static_branch_enable(&mds_idle_clear);
462 
463 	/*
464 	 * Check if the system has the right microcode.
465 	 *
466 	 * CPU Fill buffer clear mitigation is enumerated by either an explicit
467 	 * FB_CLEAR or by the presence of both MD_CLEAR and L1D_FLUSH on MDS
468 	 * affected systems.
469 	 */
470 	if ((ia32_cap & ARCH_CAP_FB_CLEAR) ||
471 	    (boot_cpu_has(X86_FEATURE_MD_CLEAR) &&
472 	     boot_cpu_has(X86_FEATURE_FLUSH_L1D) &&
473 	     !(ia32_cap & ARCH_CAP_MDS_NO)))
474 		mmio_mitigation = MMIO_MITIGATION_VERW;
475 	else
476 		mmio_mitigation = MMIO_MITIGATION_UCODE_NEEDED;
477 
478 	if (mmio_nosmt || cpu_mitigations_auto_nosmt())
479 		cpu_smt_disable(false);
480 }
481 
mmio_stale_data_parse_cmdline(char * str)482 static int __init mmio_stale_data_parse_cmdline(char *str)
483 {
484 	if (!boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA))
485 		return 0;
486 
487 	if (!str)
488 		return -EINVAL;
489 
490 	if (!strcmp(str, "off")) {
491 		mmio_mitigation = MMIO_MITIGATION_OFF;
492 	} else if (!strcmp(str, "full")) {
493 		mmio_mitigation = MMIO_MITIGATION_VERW;
494 	} else if (!strcmp(str, "full,nosmt")) {
495 		mmio_mitigation = MMIO_MITIGATION_VERW;
496 		mmio_nosmt = true;
497 	}
498 
499 	return 0;
500 }
501 early_param("mmio_stale_data", mmio_stale_data_parse_cmdline);
502 
503 #undef pr_fmt
504 #define pr_fmt(fmt)     "" fmt
505 
md_clear_update_mitigation(void)506 static void __init md_clear_update_mitigation(void)
507 {
508 	if (cpu_mitigations_off())
509 		return;
510 
511 	if (!static_key_enabled(&mds_user_clear))
512 		goto out;
513 
514 	/*
515 	 * mds_user_clear is now enabled. Update MDS, TAA and MMIO Stale Data
516 	 * mitigation, if necessary.
517 	 */
518 	if (mds_mitigation == MDS_MITIGATION_OFF &&
519 	    boot_cpu_has_bug(X86_BUG_MDS)) {
520 		mds_mitigation = MDS_MITIGATION_FULL;
521 		mds_select_mitigation();
522 	}
523 	if (taa_mitigation == TAA_MITIGATION_OFF &&
524 	    boot_cpu_has_bug(X86_BUG_TAA)) {
525 		taa_mitigation = TAA_MITIGATION_VERW;
526 		taa_select_mitigation();
527 	}
528 	if (mmio_mitigation == MMIO_MITIGATION_OFF &&
529 	    boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA)) {
530 		mmio_mitigation = MMIO_MITIGATION_VERW;
531 		mmio_select_mitigation();
532 	}
533 out:
534 	if (boot_cpu_has_bug(X86_BUG_MDS))
535 		pr_info("MDS: %s\n", mds_strings[mds_mitigation]);
536 	if (boot_cpu_has_bug(X86_BUG_TAA))
537 		pr_info("TAA: %s\n", taa_strings[taa_mitigation]);
538 	if (boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA))
539 		pr_info("MMIO Stale Data: %s\n", mmio_strings[mmio_mitigation]);
540 	else if (boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN))
541 		pr_info("MMIO Stale Data: Unknown: No mitigations\n");
542 }
543 
md_clear_select_mitigation(void)544 static void __init md_clear_select_mitigation(void)
545 {
546 	mds_select_mitigation();
547 	taa_select_mitigation();
548 	mmio_select_mitigation();
549 
550 	/*
551 	 * As MDS, TAA and MMIO Stale Data mitigations are inter-related, update
552 	 * and print their mitigation after MDS, TAA and MMIO Stale Data
553 	 * mitigation selection is done.
554 	 */
555 	md_clear_update_mitigation();
556 }
557 
558 #undef pr_fmt
559 #define pr_fmt(fmt)	"SRBDS: " fmt
560 
561 enum srbds_mitigations {
562 	SRBDS_MITIGATION_OFF,
563 	SRBDS_MITIGATION_UCODE_NEEDED,
564 	SRBDS_MITIGATION_FULL,
565 	SRBDS_MITIGATION_TSX_OFF,
566 	SRBDS_MITIGATION_HYPERVISOR,
567 };
568 
569 static enum srbds_mitigations srbds_mitigation __ro_after_init = SRBDS_MITIGATION_FULL;
570 
571 static const char * const srbds_strings[] = {
572 	[SRBDS_MITIGATION_OFF]		= "Vulnerable",
573 	[SRBDS_MITIGATION_UCODE_NEEDED]	= "Vulnerable: No microcode",
574 	[SRBDS_MITIGATION_FULL]		= "Mitigation: Microcode",
575 	[SRBDS_MITIGATION_TSX_OFF]	= "Mitigation: TSX disabled",
576 	[SRBDS_MITIGATION_HYPERVISOR]	= "Unknown: Dependent on hypervisor status",
577 };
578 
579 static bool srbds_off;
580 
update_srbds_msr(void)581 void update_srbds_msr(void)
582 {
583 	u64 mcu_ctrl;
584 
585 	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
586 		return;
587 
588 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
589 		return;
590 
591 	if (srbds_mitigation == SRBDS_MITIGATION_UCODE_NEEDED)
592 		return;
593 
594 	rdmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
595 
596 	switch (srbds_mitigation) {
597 	case SRBDS_MITIGATION_OFF:
598 	case SRBDS_MITIGATION_TSX_OFF:
599 		mcu_ctrl |= RNGDS_MITG_DIS;
600 		break;
601 	case SRBDS_MITIGATION_FULL:
602 		mcu_ctrl &= ~RNGDS_MITG_DIS;
603 		break;
604 	default:
605 		break;
606 	}
607 
608 	wrmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
609 }
610 
srbds_select_mitigation(void)611 static void __init srbds_select_mitigation(void)
612 {
613 	u64 ia32_cap;
614 
615 	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
616 		return;
617 
618 	/*
619 	 * Check to see if this is one of the MDS_NO systems supporting TSX that
620 	 * are only exposed to SRBDS when TSX is enabled or when CPU is affected
621 	 * by Processor MMIO Stale Data vulnerability.
622 	 */
623 	ia32_cap = x86_read_arch_cap_msr();
624 	if ((ia32_cap & ARCH_CAP_MDS_NO) && !boot_cpu_has(X86_FEATURE_RTM) &&
625 	    !boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA))
626 		srbds_mitigation = SRBDS_MITIGATION_TSX_OFF;
627 	else if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
628 		srbds_mitigation = SRBDS_MITIGATION_HYPERVISOR;
629 	else if (!boot_cpu_has(X86_FEATURE_SRBDS_CTRL))
630 		srbds_mitigation = SRBDS_MITIGATION_UCODE_NEEDED;
631 	else if (cpu_mitigations_off() || srbds_off)
632 		srbds_mitigation = SRBDS_MITIGATION_OFF;
633 
634 	update_srbds_msr();
635 	pr_info("%s\n", srbds_strings[srbds_mitigation]);
636 }
637 
srbds_parse_cmdline(char * str)638 static int __init srbds_parse_cmdline(char *str)
639 {
640 	if (!str)
641 		return -EINVAL;
642 
643 	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
644 		return 0;
645 
646 	srbds_off = !strcmp(str, "off");
647 	return 0;
648 }
649 early_param("srbds", srbds_parse_cmdline);
650 
651 #undef pr_fmt
652 #define pr_fmt(fmt)     "Spectre V1 : " fmt
653 
654 enum spectre_v1_mitigation {
655 	SPECTRE_V1_MITIGATION_NONE,
656 	SPECTRE_V1_MITIGATION_AUTO,
657 };
658 
659 static enum spectre_v1_mitigation spectre_v1_mitigation __ro_after_init =
660 	SPECTRE_V1_MITIGATION_AUTO;
661 
662 static const char * const spectre_v1_strings[] = {
663 	[SPECTRE_V1_MITIGATION_NONE] = "Vulnerable: __user pointer sanitization and usercopy barriers only; no swapgs barriers",
664 	[SPECTRE_V1_MITIGATION_AUTO] = "Mitigation: usercopy/swapgs barriers and __user pointer sanitization",
665 };
666 
667 /*
668  * Does SMAP provide full mitigation against speculative kernel access to
669  * userspace?
670  */
smap_works_speculatively(void)671 static bool smap_works_speculatively(void)
672 {
673 	if (!boot_cpu_has(X86_FEATURE_SMAP))
674 		return false;
675 
676 	/*
677 	 * On CPUs which are vulnerable to Meltdown, SMAP does not
678 	 * prevent speculative access to user data in the L1 cache.
679 	 * Consider SMAP to be non-functional as a mitigation on these
680 	 * CPUs.
681 	 */
682 	if (boot_cpu_has(X86_BUG_CPU_MELTDOWN))
683 		return false;
684 
685 	return true;
686 }
687 
spectre_v1_select_mitigation(void)688 static void __init spectre_v1_select_mitigation(void)
689 {
690 	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V1) || cpu_mitigations_off()) {
691 		spectre_v1_mitigation = SPECTRE_V1_MITIGATION_NONE;
692 		return;
693 	}
694 
695 	if (spectre_v1_mitigation == SPECTRE_V1_MITIGATION_AUTO) {
696 		/*
697 		 * With Spectre v1, a user can speculatively control either
698 		 * path of a conditional swapgs with a user-controlled GS
699 		 * value.  The mitigation is to add lfences to both code paths.
700 		 *
701 		 * If FSGSBASE is enabled, the user can put a kernel address in
702 		 * GS, in which case SMAP provides no protection.
703 		 *
704 		 * If FSGSBASE is disabled, the user can only put a user space
705 		 * address in GS.  That makes an attack harder, but still
706 		 * possible if there's no SMAP protection.
707 		 */
708 		if (boot_cpu_has(X86_FEATURE_FSGSBASE) ||
709 		    !smap_works_speculatively()) {
710 			/*
711 			 * Mitigation can be provided from SWAPGS itself or
712 			 * PTI as the CR3 write in the Meltdown mitigation
713 			 * is serializing.
714 			 *
715 			 * If neither is there, mitigate with an LFENCE to
716 			 * stop speculation through swapgs.
717 			 */
718 			if (boot_cpu_has_bug(X86_BUG_SWAPGS) &&
719 			    !boot_cpu_has(X86_FEATURE_PTI))
720 				setup_force_cpu_cap(X86_FEATURE_FENCE_SWAPGS_USER);
721 
722 			/*
723 			 * Enable lfences in the kernel entry (non-swapgs)
724 			 * paths, to prevent user entry from speculatively
725 			 * skipping swapgs.
726 			 */
727 			setup_force_cpu_cap(X86_FEATURE_FENCE_SWAPGS_KERNEL);
728 		}
729 	}
730 
731 	pr_info("%s\n", spectre_v1_strings[spectre_v1_mitigation]);
732 }
733 
nospectre_v1_cmdline(char * str)734 static int __init nospectre_v1_cmdline(char *str)
735 {
736 	spectre_v1_mitigation = SPECTRE_V1_MITIGATION_NONE;
737 	return 0;
738 }
739 early_param("nospectre_v1", nospectre_v1_cmdline);
740 
741 static enum spectre_v2_mitigation spectre_v2_enabled __ro_after_init =
742 	SPECTRE_V2_NONE;
743 
744 #undef pr_fmt
745 #define pr_fmt(fmt)     "RETBleed: " fmt
746 
747 enum retbleed_mitigation {
748 	RETBLEED_MITIGATION_NONE,
749 	RETBLEED_MITIGATION_UNRET,
750 	RETBLEED_MITIGATION_IBPB,
751 	RETBLEED_MITIGATION_IBRS,
752 	RETBLEED_MITIGATION_EIBRS,
753 };
754 
755 enum retbleed_mitigation_cmd {
756 	RETBLEED_CMD_OFF,
757 	RETBLEED_CMD_AUTO,
758 	RETBLEED_CMD_UNRET,
759 	RETBLEED_CMD_IBPB,
760 };
761 
762 const char * const retbleed_strings[] = {
763 	[RETBLEED_MITIGATION_NONE]	= "Vulnerable",
764 	[RETBLEED_MITIGATION_UNRET]	= "Mitigation: untrained return thunk",
765 	[RETBLEED_MITIGATION_IBPB]	= "Mitigation: IBPB",
766 	[RETBLEED_MITIGATION_IBRS]	= "Mitigation: IBRS",
767 	[RETBLEED_MITIGATION_EIBRS]	= "Mitigation: Enhanced IBRS",
768 };
769 
770 static enum retbleed_mitigation retbleed_mitigation __ro_after_init =
771 	RETBLEED_MITIGATION_NONE;
772 static enum retbleed_mitigation_cmd retbleed_cmd __ro_after_init =
773 	RETBLEED_CMD_AUTO;
774 
775 static int __ro_after_init retbleed_nosmt = false;
776 
retbleed_parse_cmdline(char * str)777 static int __init retbleed_parse_cmdline(char *str)
778 {
779 	if (!str)
780 		return -EINVAL;
781 
782 	while (str) {
783 		char *next = strchr(str, ',');
784 		if (next) {
785 			*next = 0;
786 			next++;
787 		}
788 
789 		if (!strcmp(str, "off")) {
790 			retbleed_cmd = RETBLEED_CMD_OFF;
791 		} else if (!strcmp(str, "auto")) {
792 			retbleed_cmd = RETBLEED_CMD_AUTO;
793 		} else if (!strcmp(str, "unret")) {
794 			retbleed_cmd = RETBLEED_CMD_UNRET;
795 		} else if (!strcmp(str, "ibpb")) {
796 			retbleed_cmd = RETBLEED_CMD_IBPB;
797 		} else if (!strcmp(str, "nosmt")) {
798 			retbleed_nosmt = true;
799 		} else {
800 			pr_err("Ignoring unknown retbleed option (%s).", str);
801 		}
802 
803 		str = next;
804 	}
805 
806 	return 0;
807 }
808 early_param("retbleed", retbleed_parse_cmdline);
809 
810 #define RETBLEED_UNTRAIN_MSG "WARNING: BTB untrained return thunk mitigation is only effective on AMD/Hygon!\n"
811 #define RETBLEED_INTEL_MSG "WARNING: Spectre v2 mitigation leaves CPU vulnerable to RETBleed attacks, data leaks possible!\n"
812 
retbleed_select_mitigation(void)813 static void __init retbleed_select_mitigation(void)
814 {
815 	bool mitigate_smt = false;
816 
817 	if (!boot_cpu_has_bug(X86_BUG_RETBLEED) || cpu_mitigations_off())
818 		return;
819 
820 	switch (retbleed_cmd) {
821 	case RETBLEED_CMD_OFF:
822 		return;
823 
824 	case RETBLEED_CMD_UNRET:
825 		if (IS_ENABLED(CONFIG_CPU_UNRET_ENTRY)) {
826 			retbleed_mitigation = RETBLEED_MITIGATION_UNRET;
827 		} else {
828 			pr_err("WARNING: kernel not compiled with CPU_UNRET_ENTRY.\n");
829 			goto do_cmd_auto;
830 		}
831 		break;
832 
833 	case RETBLEED_CMD_IBPB:
834 		if (!boot_cpu_has(X86_FEATURE_IBPB)) {
835 			pr_err("WARNING: CPU does not support IBPB.\n");
836 			goto do_cmd_auto;
837 		} else if (IS_ENABLED(CONFIG_CPU_IBPB_ENTRY)) {
838 			retbleed_mitigation = RETBLEED_MITIGATION_IBPB;
839 		} else {
840 			pr_err("WARNING: kernel not compiled with CPU_IBPB_ENTRY.\n");
841 			goto do_cmd_auto;
842 		}
843 		break;
844 
845 do_cmd_auto:
846 	case RETBLEED_CMD_AUTO:
847 	default:
848 		if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
849 		    boot_cpu_data.x86_vendor == X86_VENDOR_HYGON) {
850 			if (IS_ENABLED(CONFIG_CPU_UNRET_ENTRY))
851 				retbleed_mitigation = RETBLEED_MITIGATION_UNRET;
852 			else if (IS_ENABLED(CONFIG_CPU_IBPB_ENTRY) && boot_cpu_has(X86_FEATURE_IBPB))
853 				retbleed_mitigation = RETBLEED_MITIGATION_IBPB;
854 		}
855 
856 		/*
857 		 * The Intel mitigation (IBRS or eIBRS) was already selected in
858 		 * spectre_v2_select_mitigation().  'retbleed_mitigation' will
859 		 * be set accordingly below.
860 		 */
861 
862 		break;
863 	}
864 
865 	switch (retbleed_mitigation) {
866 	case RETBLEED_MITIGATION_UNRET:
867 		setup_force_cpu_cap(X86_FEATURE_RETHUNK);
868 		setup_force_cpu_cap(X86_FEATURE_UNRET);
869 
870 		if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD &&
871 		    boot_cpu_data.x86_vendor != X86_VENDOR_HYGON)
872 			pr_err(RETBLEED_UNTRAIN_MSG);
873 
874 		mitigate_smt = true;
875 		break;
876 
877 	case RETBLEED_MITIGATION_IBPB:
878 		setup_force_cpu_cap(X86_FEATURE_ENTRY_IBPB);
879 		mitigate_smt = true;
880 		break;
881 
882 	default:
883 		break;
884 	}
885 
886 	if (mitigate_smt && !boot_cpu_has(X86_FEATURE_STIBP) &&
887 	    (retbleed_nosmt || cpu_mitigations_auto_nosmt()))
888 		cpu_smt_disable(false);
889 
890 	/*
891 	 * Let IBRS trump all on Intel without affecting the effects of the
892 	 * retbleed= cmdline option.
893 	 */
894 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) {
895 		switch (spectre_v2_enabled) {
896 		case SPECTRE_V2_IBRS:
897 			retbleed_mitigation = RETBLEED_MITIGATION_IBRS;
898 			break;
899 		case SPECTRE_V2_EIBRS:
900 		case SPECTRE_V2_EIBRS_RETPOLINE:
901 		case SPECTRE_V2_EIBRS_LFENCE:
902 			retbleed_mitigation = RETBLEED_MITIGATION_EIBRS;
903 			break;
904 		default:
905 			pr_err(RETBLEED_INTEL_MSG);
906 		}
907 	}
908 
909 	pr_info("%s\n", retbleed_strings[retbleed_mitigation]);
910 }
911 
912 #undef pr_fmt
913 #define pr_fmt(fmt)     "Spectre V2 : " fmt
914 
915 static enum spectre_v2_user_mitigation spectre_v2_user_stibp __ro_after_init =
916 	SPECTRE_V2_USER_NONE;
917 static enum spectre_v2_user_mitigation spectre_v2_user_ibpb __ro_after_init =
918 	SPECTRE_V2_USER_NONE;
919 
920 #ifdef CONFIG_RETPOLINE
921 static bool spectre_v2_bad_module;
922 
retpoline_module_ok(bool has_retpoline)923 bool retpoline_module_ok(bool has_retpoline)
924 {
925 	if (spectre_v2_enabled == SPECTRE_V2_NONE || has_retpoline)
926 		return true;
927 
928 	pr_err("System may be vulnerable to spectre v2\n");
929 	spectre_v2_bad_module = true;
930 	return false;
931 }
932 
spectre_v2_module_string(void)933 static inline const char *spectre_v2_module_string(void)
934 {
935 	return spectre_v2_bad_module ? " - vulnerable module loaded" : "";
936 }
937 #else
spectre_v2_module_string(void)938 static inline const char *spectre_v2_module_string(void) { return ""; }
939 #endif
940 
941 #define SPECTRE_V2_LFENCE_MSG "WARNING: LFENCE mitigation is not recommended for this CPU, data leaks possible!\n"
942 #define SPECTRE_V2_EIBRS_EBPF_MSG "WARNING: Unprivileged eBPF is enabled with eIBRS on, data leaks possible via Spectre v2 BHB attacks!\n"
943 #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"
944 #define SPECTRE_V2_IBRS_PERF_MSG "WARNING: IBRS mitigation selected on Enhanced IBRS CPU, this may cause unnecessary performance loss\n"
945 
946 #ifdef CONFIG_BPF_SYSCALL
unpriv_ebpf_notify(int new_state)947 void unpriv_ebpf_notify(int new_state)
948 {
949 	if (new_state)
950 		return;
951 
952 	/* Unprivileged eBPF is enabled */
953 
954 	switch (spectre_v2_enabled) {
955 	case SPECTRE_V2_EIBRS:
956 		pr_err(SPECTRE_V2_EIBRS_EBPF_MSG);
957 		break;
958 	case SPECTRE_V2_EIBRS_LFENCE:
959 		if (sched_smt_active())
960 			pr_err(SPECTRE_V2_EIBRS_LFENCE_EBPF_SMT_MSG);
961 		break;
962 	default:
963 		break;
964 	}
965 }
966 #endif
967 
match_option(const char * arg,int arglen,const char * opt)968 static inline bool match_option(const char *arg, int arglen, const char *opt)
969 {
970 	int len = strlen(opt);
971 
972 	return len == arglen && !strncmp(arg, opt, len);
973 }
974 
975 /* The kernel command line selection for spectre v2 */
976 enum spectre_v2_mitigation_cmd {
977 	SPECTRE_V2_CMD_NONE,
978 	SPECTRE_V2_CMD_AUTO,
979 	SPECTRE_V2_CMD_FORCE,
980 	SPECTRE_V2_CMD_RETPOLINE,
981 	SPECTRE_V2_CMD_RETPOLINE_GENERIC,
982 	SPECTRE_V2_CMD_RETPOLINE_LFENCE,
983 	SPECTRE_V2_CMD_EIBRS,
984 	SPECTRE_V2_CMD_EIBRS_RETPOLINE,
985 	SPECTRE_V2_CMD_EIBRS_LFENCE,
986 	SPECTRE_V2_CMD_IBRS,
987 };
988 
989 enum spectre_v2_user_cmd {
990 	SPECTRE_V2_USER_CMD_NONE,
991 	SPECTRE_V2_USER_CMD_AUTO,
992 	SPECTRE_V2_USER_CMD_FORCE,
993 	SPECTRE_V2_USER_CMD_PRCTL,
994 	SPECTRE_V2_USER_CMD_PRCTL_IBPB,
995 	SPECTRE_V2_USER_CMD_SECCOMP,
996 	SPECTRE_V2_USER_CMD_SECCOMP_IBPB,
997 };
998 
999 static const char * const spectre_v2_user_strings[] = {
1000 	[SPECTRE_V2_USER_NONE]			= "User space: Vulnerable",
1001 	[SPECTRE_V2_USER_STRICT]		= "User space: Mitigation: STIBP protection",
1002 	[SPECTRE_V2_USER_STRICT_PREFERRED]	= "User space: Mitigation: STIBP always-on protection",
1003 	[SPECTRE_V2_USER_PRCTL]			= "User space: Mitigation: STIBP via prctl",
1004 	[SPECTRE_V2_USER_SECCOMP]		= "User space: Mitigation: STIBP via seccomp and prctl",
1005 };
1006 
1007 static const struct {
1008 	const char			*option;
1009 	enum spectre_v2_user_cmd	cmd;
1010 	bool				secure;
1011 } v2_user_options[] __initconst = {
1012 	{ "auto",		SPECTRE_V2_USER_CMD_AUTO,		false },
1013 	{ "off",		SPECTRE_V2_USER_CMD_NONE,		false },
1014 	{ "on",			SPECTRE_V2_USER_CMD_FORCE,		true  },
1015 	{ "prctl",		SPECTRE_V2_USER_CMD_PRCTL,		false },
1016 	{ "prctl,ibpb",		SPECTRE_V2_USER_CMD_PRCTL_IBPB,		false },
1017 	{ "seccomp",		SPECTRE_V2_USER_CMD_SECCOMP,		false },
1018 	{ "seccomp,ibpb",	SPECTRE_V2_USER_CMD_SECCOMP_IBPB,	false },
1019 };
1020 
spec_v2_user_print_cond(const char * reason,bool secure)1021 static void __init spec_v2_user_print_cond(const char *reason, bool secure)
1022 {
1023 	if (boot_cpu_has_bug(X86_BUG_SPECTRE_V2) != secure)
1024 		pr_info("spectre_v2_user=%s forced on command line.\n", reason);
1025 }
1026 
1027 static __ro_after_init enum spectre_v2_mitigation_cmd spectre_v2_cmd;
1028 
1029 static enum spectre_v2_user_cmd __init
spectre_v2_parse_user_cmdline(void)1030 spectre_v2_parse_user_cmdline(void)
1031 {
1032 	char arg[20];
1033 	int ret, i;
1034 
1035 	switch (spectre_v2_cmd) {
1036 	case SPECTRE_V2_CMD_NONE:
1037 		return SPECTRE_V2_USER_CMD_NONE;
1038 	case SPECTRE_V2_CMD_FORCE:
1039 		return SPECTRE_V2_USER_CMD_FORCE;
1040 	default:
1041 		break;
1042 	}
1043 
1044 	ret = cmdline_find_option(boot_command_line, "spectre_v2_user",
1045 				  arg, sizeof(arg));
1046 	if (ret < 0)
1047 		return SPECTRE_V2_USER_CMD_AUTO;
1048 
1049 	for (i = 0; i < ARRAY_SIZE(v2_user_options); i++) {
1050 		if (match_option(arg, ret, v2_user_options[i].option)) {
1051 			spec_v2_user_print_cond(v2_user_options[i].option,
1052 						v2_user_options[i].secure);
1053 			return v2_user_options[i].cmd;
1054 		}
1055 	}
1056 
1057 	pr_err("Unknown user space protection option (%s). Switching to AUTO select\n", arg);
1058 	return SPECTRE_V2_USER_CMD_AUTO;
1059 }
1060 
spectre_v2_in_eibrs_mode(enum spectre_v2_mitigation mode)1061 static inline bool spectre_v2_in_eibrs_mode(enum spectre_v2_mitigation mode)
1062 {
1063 	return mode == SPECTRE_V2_EIBRS ||
1064 	       mode == SPECTRE_V2_EIBRS_RETPOLINE ||
1065 	       mode == SPECTRE_V2_EIBRS_LFENCE;
1066 }
1067 
spectre_v2_in_ibrs_mode(enum spectre_v2_mitigation mode)1068 static inline bool spectre_v2_in_ibrs_mode(enum spectre_v2_mitigation mode)
1069 {
1070 	return spectre_v2_in_eibrs_mode(mode) || mode == SPECTRE_V2_IBRS;
1071 }
1072 
1073 static void __init
spectre_v2_user_select_mitigation(void)1074 spectre_v2_user_select_mitigation(void)
1075 {
1076 	enum spectre_v2_user_mitigation mode = SPECTRE_V2_USER_NONE;
1077 	bool smt_possible = IS_ENABLED(CONFIG_SMP);
1078 	enum spectre_v2_user_cmd cmd;
1079 
1080 	if (!boot_cpu_has(X86_FEATURE_IBPB) && !boot_cpu_has(X86_FEATURE_STIBP))
1081 		return;
1082 
1083 	if (cpu_smt_control == CPU_SMT_FORCE_DISABLED ||
1084 	    cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
1085 		smt_possible = false;
1086 
1087 	cmd = spectre_v2_parse_user_cmdline();
1088 	switch (cmd) {
1089 	case SPECTRE_V2_USER_CMD_NONE:
1090 		goto set_mode;
1091 	case SPECTRE_V2_USER_CMD_FORCE:
1092 		mode = SPECTRE_V2_USER_STRICT;
1093 		break;
1094 	case SPECTRE_V2_USER_CMD_PRCTL:
1095 	case SPECTRE_V2_USER_CMD_PRCTL_IBPB:
1096 		mode = SPECTRE_V2_USER_PRCTL;
1097 		break;
1098 	case SPECTRE_V2_USER_CMD_AUTO:
1099 	case SPECTRE_V2_USER_CMD_SECCOMP:
1100 	case SPECTRE_V2_USER_CMD_SECCOMP_IBPB:
1101 		if (IS_ENABLED(CONFIG_SECCOMP))
1102 			mode = SPECTRE_V2_USER_SECCOMP;
1103 		else
1104 			mode = SPECTRE_V2_USER_PRCTL;
1105 		break;
1106 	}
1107 
1108 	/* Initialize Indirect Branch Prediction Barrier */
1109 	if (boot_cpu_has(X86_FEATURE_IBPB)) {
1110 		setup_force_cpu_cap(X86_FEATURE_USE_IBPB);
1111 
1112 		spectre_v2_user_ibpb = mode;
1113 		switch (cmd) {
1114 		case SPECTRE_V2_USER_CMD_FORCE:
1115 		case SPECTRE_V2_USER_CMD_PRCTL_IBPB:
1116 		case SPECTRE_V2_USER_CMD_SECCOMP_IBPB:
1117 			static_branch_enable(&switch_mm_always_ibpb);
1118 			spectre_v2_user_ibpb = SPECTRE_V2_USER_STRICT;
1119 			break;
1120 		case SPECTRE_V2_USER_CMD_PRCTL:
1121 		case SPECTRE_V2_USER_CMD_AUTO:
1122 		case SPECTRE_V2_USER_CMD_SECCOMP:
1123 			static_branch_enable(&switch_mm_cond_ibpb);
1124 			break;
1125 		default:
1126 			break;
1127 		}
1128 
1129 		pr_info("mitigation: Enabling %s Indirect Branch Prediction Barrier\n",
1130 			static_key_enabled(&switch_mm_always_ibpb) ?
1131 			"always-on" : "conditional");
1132 	}
1133 
1134 	/*
1135 	 * If no STIBP, enhanced IBRS is enabled, or SMT impossible, STIBP
1136 	 * is not required.
1137 	 *
1138 	 * Enhanced IBRS also protects against cross-thread branch target
1139 	 * injection in user-mode as the IBRS bit remains always set which
1140 	 * implicitly enables cross-thread protections.  However, in legacy IBRS
1141 	 * mode, the IBRS bit is set only on kernel entry and cleared on return
1142 	 * to userspace. This disables the implicit cross-thread protection,
1143 	 * so allow for STIBP to be selected in that case.
1144 	 */
1145 	if (!boot_cpu_has(X86_FEATURE_STIBP) ||
1146 	    !smt_possible ||
1147 	    spectre_v2_in_eibrs_mode(spectre_v2_enabled))
1148 		return;
1149 
1150 	/*
1151 	 * At this point, an STIBP mode other than "off" has been set.
1152 	 * If STIBP support is not being forced, check if STIBP always-on
1153 	 * is preferred.
1154 	 */
1155 	if (mode != SPECTRE_V2_USER_STRICT &&
1156 	    boot_cpu_has(X86_FEATURE_AMD_STIBP_ALWAYS_ON))
1157 		mode = SPECTRE_V2_USER_STRICT_PREFERRED;
1158 
1159 	if (retbleed_mitigation == RETBLEED_MITIGATION_UNRET ||
1160 	    retbleed_mitigation == RETBLEED_MITIGATION_IBPB) {
1161 		if (mode != SPECTRE_V2_USER_STRICT &&
1162 		    mode != SPECTRE_V2_USER_STRICT_PREFERRED)
1163 			pr_info("Selecting STIBP always-on mode to complement retbleed mitigation\n");
1164 		mode = SPECTRE_V2_USER_STRICT_PREFERRED;
1165 	}
1166 
1167 	spectre_v2_user_stibp = mode;
1168 
1169 set_mode:
1170 	pr_info("%s\n", spectre_v2_user_strings[mode]);
1171 }
1172 
1173 static const char * const spectre_v2_strings[] = {
1174 	[SPECTRE_V2_NONE]			= "Vulnerable",
1175 	[SPECTRE_V2_RETPOLINE]			= "Mitigation: Retpolines",
1176 	[SPECTRE_V2_LFENCE]			= "Mitigation: LFENCE",
1177 	[SPECTRE_V2_EIBRS]			= "Mitigation: Enhanced IBRS",
1178 	[SPECTRE_V2_EIBRS_LFENCE]		= "Mitigation: Enhanced IBRS + LFENCE",
1179 	[SPECTRE_V2_EIBRS_RETPOLINE]		= "Mitigation: Enhanced IBRS + Retpolines",
1180 	[SPECTRE_V2_IBRS]			= "Mitigation: IBRS",
1181 };
1182 
1183 static const struct {
1184 	const char *option;
1185 	enum spectre_v2_mitigation_cmd cmd;
1186 	bool secure;
1187 } mitigation_options[] __initconst = {
1188 	{ "off",		SPECTRE_V2_CMD_NONE,		  false },
1189 	{ "on",			SPECTRE_V2_CMD_FORCE,		  true  },
1190 	{ "retpoline",		SPECTRE_V2_CMD_RETPOLINE,	  false },
1191 	{ "retpoline,amd",	SPECTRE_V2_CMD_RETPOLINE_LFENCE,  false },
1192 	{ "retpoline,lfence",	SPECTRE_V2_CMD_RETPOLINE_LFENCE,  false },
1193 	{ "retpoline,generic",	SPECTRE_V2_CMD_RETPOLINE_GENERIC, false },
1194 	{ "eibrs",		SPECTRE_V2_CMD_EIBRS,		  false },
1195 	{ "eibrs,lfence",	SPECTRE_V2_CMD_EIBRS_LFENCE,	  false },
1196 	{ "eibrs,retpoline",	SPECTRE_V2_CMD_EIBRS_RETPOLINE,	  false },
1197 	{ "auto",		SPECTRE_V2_CMD_AUTO,		  false },
1198 	{ "ibrs",		SPECTRE_V2_CMD_IBRS,              false },
1199 };
1200 
spec_v2_print_cond(const char * reason,bool secure)1201 static void __init spec_v2_print_cond(const char *reason, bool secure)
1202 {
1203 	if (boot_cpu_has_bug(X86_BUG_SPECTRE_V2) != secure)
1204 		pr_info("%s selected on command line.\n", reason);
1205 }
1206 
spectre_v2_parse_cmdline(void)1207 static enum spectre_v2_mitigation_cmd __init spectre_v2_parse_cmdline(void)
1208 {
1209 	enum spectre_v2_mitigation_cmd cmd = SPECTRE_V2_CMD_AUTO;
1210 	char arg[20];
1211 	int ret, i;
1212 
1213 	if (cmdline_find_option_bool(boot_command_line, "nospectre_v2") ||
1214 	    cpu_mitigations_off())
1215 		return SPECTRE_V2_CMD_NONE;
1216 
1217 	ret = cmdline_find_option(boot_command_line, "spectre_v2", arg, sizeof(arg));
1218 	if (ret < 0)
1219 		return SPECTRE_V2_CMD_AUTO;
1220 
1221 	for (i = 0; i < ARRAY_SIZE(mitigation_options); i++) {
1222 		if (!match_option(arg, ret, mitigation_options[i].option))
1223 			continue;
1224 		cmd = mitigation_options[i].cmd;
1225 		break;
1226 	}
1227 
1228 	if (i >= ARRAY_SIZE(mitigation_options)) {
1229 		pr_err("unknown option (%s). Switching to AUTO select\n", arg);
1230 		return SPECTRE_V2_CMD_AUTO;
1231 	}
1232 
1233 	if ((cmd == SPECTRE_V2_CMD_RETPOLINE ||
1234 	     cmd == SPECTRE_V2_CMD_RETPOLINE_LFENCE ||
1235 	     cmd == SPECTRE_V2_CMD_RETPOLINE_GENERIC ||
1236 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE ||
1237 	     cmd == SPECTRE_V2_CMD_EIBRS_RETPOLINE) &&
1238 	    !IS_ENABLED(CONFIG_RETPOLINE)) {
1239 		pr_err("%s selected but not compiled in. Switching to AUTO select\n",
1240 		       mitigation_options[i].option);
1241 		return SPECTRE_V2_CMD_AUTO;
1242 	}
1243 
1244 	if ((cmd == SPECTRE_V2_CMD_EIBRS ||
1245 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE ||
1246 	     cmd == SPECTRE_V2_CMD_EIBRS_RETPOLINE) &&
1247 	    !boot_cpu_has(X86_FEATURE_IBRS_ENHANCED)) {
1248 		pr_err("%s selected but CPU doesn't have eIBRS. Switching to AUTO select\n",
1249 		       mitigation_options[i].option);
1250 		return SPECTRE_V2_CMD_AUTO;
1251 	}
1252 
1253 	if ((cmd == SPECTRE_V2_CMD_RETPOLINE_LFENCE ||
1254 	     cmd == SPECTRE_V2_CMD_EIBRS_LFENCE) &&
1255 	    !boot_cpu_has(X86_FEATURE_LFENCE_RDTSC)) {
1256 		pr_err("%s selected, but CPU doesn't have a serializing LFENCE. Switching to AUTO select\n",
1257 		       mitigation_options[i].option);
1258 		return SPECTRE_V2_CMD_AUTO;
1259 	}
1260 
1261 	if (cmd == SPECTRE_V2_CMD_IBRS && !IS_ENABLED(CONFIG_CPU_IBRS_ENTRY)) {
1262 		pr_err("%s selected but not compiled in. Switching to AUTO select\n",
1263 		       mitigation_options[i].option);
1264 		return SPECTRE_V2_CMD_AUTO;
1265 	}
1266 
1267 	if (cmd == SPECTRE_V2_CMD_IBRS && boot_cpu_data.x86_vendor != X86_VENDOR_INTEL) {
1268 		pr_err("%s selected but not Intel CPU. Switching to AUTO select\n",
1269 		       mitigation_options[i].option);
1270 		return SPECTRE_V2_CMD_AUTO;
1271 	}
1272 
1273 	if (cmd == SPECTRE_V2_CMD_IBRS && !boot_cpu_has(X86_FEATURE_IBRS)) {
1274 		pr_err("%s selected but CPU doesn't have IBRS. Switching to AUTO select\n",
1275 		       mitigation_options[i].option);
1276 		return SPECTRE_V2_CMD_AUTO;
1277 	}
1278 
1279 	if (cmd == SPECTRE_V2_CMD_IBRS && boot_cpu_has(X86_FEATURE_XENPV)) {
1280 		pr_err("%s selected but running as XenPV guest. Switching to AUTO select\n",
1281 		       mitigation_options[i].option);
1282 		return SPECTRE_V2_CMD_AUTO;
1283 	}
1284 
1285 	spec_v2_print_cond(mitigation_options[i].option,
1286 			   mitigation_options[i].secure);
1287 	return cmd;
1288 }
1289 
spectre_v2_select_retpoline(void)1290 static enum spectre_v2_mitigation __init spectre_v2_select_retpoline(void)
1291 {
1292 	if (!IS_ENABLED(CONFIG_RETPOLINE)) {
1293 		pr_err("Kernel not compiled with retpoline; no mitigation available!");
1294 		return SPECTRE_V2_NONE;
1295 	}
1296 
1297 	return SPECTRE_V2_RETPOLINE;
1298 }
1299 
1300 /* Disable in-kernel use of non-RSB RET predictors */
spec_ctrl_disable_kernel_rrsba(void)1301 static void __init spec_ctrl_disable_kernel_rrsba(void)
1302 {
1303 	u64 ia32_cap;
1304 
1305 	if (!boot_cpu_has(X86_FEATURE_RRSBA_CTRL))
1306 		return;
1307 
1308 	ia32_cap = x86_read_arch_cap_msr();
1309 
1310 	if (ia32_cap & ARCH_CAP_RRSBA) {
1311 		x86_spec_ctrl_base |= SPEC_CTRL_RRSBA_DIS_S;
1312 		update_spec_ctrl(x86_spec_ctrl_base);
1313 	}
1314 }
1315 
spectre_v2_determine_rsb_fill_type_at_vmexit(enum spectre_v2_mitigation mode)1316 static void __init spectre_v2_determine_rsb_fill_type_at_vmexit(enum spectre_v2_mitigation mode)
1317 {
1318 	/*
1319 	 * Similar to context switches, there are two types of RSB attacks
1320 	 * after VM exit:
1321 	 *
1322 	 * 1) RSB underflow
1323 	 *
1324 	 * 2) Poisoned RSB entry
1325 	 *
1326 	 * When retpoline is enabled, both are mitigated by filling/clearing
1327 	 * the RSB.
1328 	 *
1329 	 * When IBRS is enabled, while #1 would be mitigated by the IBRS branch
1330 	 * prediction isolation protections, RSB still needs to be cleared
1331 	 * because of #2.  Note that SMEP provides no protection here, unlike
1332 	 * user-space-poisoned RSB entries.
1333 	 *
1334 	 * eIBRS should protect against RSB poisoning, but if the EIBRS_PBRSB
1335 	 * bug is present then a LITE version of RSB protection is required,
1336 	 * just a single call needs to retire before a RET is executed.
1337 	 */
1338 	switch (mode) {
1339 	case SPECTRE_V2_NONE:
1340 		return;
1341 
1342 	case SPECTRE_V2_EIBRS_LFENCE:
1343 	case SPECTRE_V2_EIBRS:
1344 		if (boot_cpu_has_bug(X86_BUG_EIBRS_PBRSB)) {
1345 			setup_force_cpu_cap(X86_FEATURE_RSB_VMEXIT_LITE);
1346 			pr_info("Spectre v2 / PBRSB-eIBRS: Retire a single CALL on VMEXIT\n");
1347 		}
1348 		return;
1349 
1350 	case SPECTRE_V2_EIBRS_RETPOLINE:
1351 	case SPECTRE_V2_RETPOLINE:
1352 	case SPECTRE_V2_LFENCE:
1353 	case SPECTRE_V2_IBRS:
1354 		setup_force_cpu_cap(X86_FEATURE_RSB_VMEXIT);
1355 		pr_info("Spectre v2 / SpectreRSB : Filling RSB on VMEXIT\n");
1356 		return;
1357 	}
1358 
1359 	pr_warn_once("Unknown Spectre v2 mode, disabling RSB mitigation at VM exit");
1360 	dump_stack();
1361 }
1362 
spectre_v2_select_mitigation(void)1363 static void __init spectre_v2_select_mitigation(void)
1364 {
1365 	enum spectre_v2_mitigation_cmd cmd = spectre_v2_parse_cmdline();
1366 	enum spectre_v2_mitigation mode = SPECTRE_V2_NONE;
1367 
1368 	/*
1369 	 * If the CPU is not affected and the command line mode is NONE or AUTO
1370 	 * then nothing to do.
1371 	 */
1372 	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V2) &&
1373 	    (cmd == SPECTRE_V2_CMD_NONE || cmd == SPECTRE_V2_CMD_AUTO))
1374 		return;
1375 
1376 	switch (cmd) {
1377 	case SPECTRE_V2_CMD_NONE:
1378 		return;
1379 
1380 	case SPECTRE_V2_CMD_FORCE:
1381 	case SPECTRE_V2_CMD_AUTO:
1382 		if (boot_cpu_has(X86_FEATURE_IBRS_ENHANCED)) {
1383 			mode = SPECTRE_V2_EIBRS;
1384 			break;
1385 		}
1386 
1387 		if (IS_ENABLED(CONFIG_CPU_IBRS_ENTRY) &&
1388 		    boot_cpu_has_bug(X86_BUG_RETBLEED) &&
1389 		    retbleed_cmd != RETBLEED_CMD_OFF &&
1390 		    boot_cpu_has(X86_FEATURE_IBRS) &&
1391 		    boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) {
1392 			mode = SPECTRE_V2_IBRS;
1393 			break;
1394 		}
1395 
1396 		mode = spectre_v2_select_retpoline();
1397 		break;
1398 
1399 	case SPECTRE_V2_CMD_RETPOLINE_LFENCE:
1400 		pr_err(SPECTRE_V2_LFENCE_MSG);
1401 		mode = SPECTRE_V2_LFENCE;
1402 		break;
1403 
1404 	case SPECTRE_V2_CMD_RETPOLINE_GENERIC:
1405 		mode = SPECTRE_V2_RETPOLINE;
1406 		break;
1407 
1408 	case SPECTRE_V2_CMD_RETPOLINE:
1409 		mode = spectre_v2_select_retpoline();
1410 		break;
1411 
1412 	case SPECTRE_V2_CMD_IBRS:
1413 		mode = SPECTRE_V2_IBRS;
1414 		break;
1415 
1416 	case SPECTRE_V2_CMD_EIBRS:
1417 		mode = SPECTRE_V2_EIBRS;
1418 		break;
1419 
1420 	case SPECTRE_V2_CMD_EIBRS_LFENCE:
1421 		mode = SPECTRE_V2_EIBRS_LFENCE;
1422 		break;
1423 
1424 	case SPECTRE_V2_CMD_EIBRS_RETPOLINE:
1425 		mode = SPECTRE_V2_EIBRS_RETPOLINE;
1426 		break;
1427 	}
1428 
1429 	if (mode == SPECTRE_V2_EIBRS && unprivileged_ebpf_enabled())
1430 		pr_err(SPECTRE_V2_EIBRS_EBPF_MSG);
1431 
1432 	if (spectre_v2_in_ibrs_mode(mode)) {
1433 		x86_spec_ctrl_base |= SPEC_CTRL_IBRS;
1434 		update_spec_ctrl(x86_spec_ctrl_base);
1435 	}
1436 
1437 	switch (mode) {
1438 	case SPECTRE_V2_NONE:
1439 	case SPECTRE_V2_EIBRS:
1440 		break;
1441 
1442 	case SPECTRE_V2_IBRS:
1443 		setup_force_cpu_cap(X86_FEATURE_KERNEL_IBRS);
1444 		if (boot_cpu_has(X86_FEATURE_IBRS_ENHANCED))
1445 			pr_warn(SPECTRE_V2_IBRS_PERF_MSG);
1446 		break;
1447 
1448 	case SPECTRE_V2_LFENCE:
1449 	case SPECTRE_V2_EIBRS_LFENCE:
1450 		setup_force_cpu_cap(X86_FEATURE_RETPOLINE_LFENCE);
1451 		fallthrough;
1452 
1453 	case SPECTRE_V2_RETPOLINE:
1454 	case SPECTRE_V2_EIBRS_RETPOLINE:
1455 		setup_force_cpu_cap(X86_FEATURE_RETPOLINE);
1456 		break;
1457 	}
1458 
1459 	/*
1460 	 * Disable alternate RSB predictions in kernel when indirect CALLs and
1461 	 * JMPs gets protection against BHI and Intramode-BTI, but RET
1462 	 * prediction from a non-RSB predictor is still a risk.
1463 	 */
1464 	if (mode == SPECTRE_V2_EIBRS_LFENCE ||
1465 	    mode == SPECTRE_V2_EIBRS_RETPOLINE ||
1466 	    mode == SPECTRE_V2_RETPOLINE)
1467 		spec_ctrl_disable_kernel_rrsba();
1468 
1469 	spectre_v2_enabled = mode;
1470 	pr_info("%s\n", spectre_v2_strings[mode]);
1471 
1472 	/*
1473 	 * If Spectre v2 protection has been enabled, fill the RSB during a
1474 	 * context switch.  In general there are two types of RSB attacks
1475 	 * across context switches, for which the CALLs/RETs may be unbalanced.
1476 	 *
1477 	 * 1) RSB underflow
1478 	 *
1479 	 *    Some Intel parts have "bottomless RSB".  When the RSB is empty,
1480 	 *    speculated return targets may come from the branch predictor,
1481 	 *    which could have a user-poisoned BTB or BHB entry.
1482 	 *
1483 	 *    AMD has it even worse: *all* returns are speculated from the BTB,
1484 	 *    regardless of the state of the RSB.
1485 	 *
1486 	 *    When IBRS or eIBRS is enabled, the "user -> kernel" attack
1487 	 *    scenario is mitigated by the IBRS branch prediction isolation
1488 	 *    properties, so the RSB buffer filling wouldn't be necessary to
1489 	 *    protect against this type of attack.
1490 	 *
1491 	 *    The "user -> user" attack scenario is mitigated by RSB filling.
1492 	 *
1493 	 * 2) Poisoned RSB entry
1494 	 *
1495 	 *    If the 'next' in-kernel return stack is shorter than 'prev',
1496 	 *    'next' could be tricked into speculating with a user-poisoned RSB
1497 	 *    entry.
1498 	 *
1499 	 *    The "user -> kernel" attack scenario is mitigated by SMEP and
1500 	 *    eIBRS.
1501 	 *
1502 	 *    The "user -> user" scenario, also known as SpectreBHB, requires
1503 	 *    RSB clearing.
1504 	 *
1505 	 * So to mitigate all cases, unconditionally fill RSB on context
1506 	 * switches.
1507 	 *
1508 	 * FIXME: Is this pointless for retbleed-affected AMD?
1509 	 */
1510 	setup_force_cpu_cap(X86_FEATURE_RSB_CTXSW);
1511 	pr_info("Spectre v2 / SpectreRSB mitigation: Filling RSB on context switch\n");
1512 
1513 	spectre_v2_determine_rsb_fill_type_at_vmexit(mode);
1514 
1515 	/*
1516 	 * Retpoline protects the kernel, but doesn't protect firmware.  IBRS
1517 	 * and Enhanced IBRS protect firmware too, so enable IBRS around
1518 	 * firmware calls only when IBRS / Enhanced IBRS aren't otherwise
1519 	 * enabled.
1520 	 *
1521 	 * Use "mode" to check Enhanced IBRS instead of boot_cpu_has(), because
1522 	 * the user might select retpoline on the kernel command line and if
1523 	 * the CPU supports Enhanced IBRS, kernel might un-intentionally not
1524 	 * enable IBRS around firmware calls.
1525 	 */
1526 	if (boot_cpu_has_bug(X86_BUG_RETBLEED) &&
1527 	    boot_cpu_has(X86_FEATURE_IBPB) &&
1528 	    (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
1529 	     boot_cpu_data.x86_vendor == X86_VENDOR_HYGON)) {
1530 
1531 		if (retbleed_cmd != RETBLEED_CMD_IBPB) {
1532 			setup_force_cpu_cap(X86_FEATURE_USE_IBPB_FW);
1533 			pr_info("Enabling Speculation Barrier for firmware calls\n");
1534 		}
1535 
1536 	} else if (boot_cpu_has(X86_FEATURE_IBRS) && !spectre_v2_in_ibrs_mode(mode)) {
1537 		setup_force_cpu_cap(X86_FEATURE_USE_IBRS_FW);
1538 		pr_info("Enabling Restricted Speculation for firmware calls\n");
1539 	}
1540 
1541 	/* Set up IBPB and STIBP depending on the general spectre V2 command */
1542 	spectre_v2_cmd = cmd;
1543 }
1544 
update_stibp_msr(void * __unused)1545 static void update_stibp_msr(void * __unused)
1546 {
1547 	u64 val = spec_ctrl_current() | (x86_spec_ctrl_base & SPEC_CTRL_STIBP);
1548 	update_spec_ctrl(val);
1549 }
1550 
1551 /* Update x86_spec_ctrl_base in case SMT state changed. */
update_stibp_strict(void)1552 static void update_stibp_strict(void)
1553 {
1554 	u64 mask = x86_spec_ctrl_base & ~SPEC_CTRL_STIBP;
1555 
1556 	if (sched_smt_active())
1557 		mask |= SPEC_CTRL_STIBP;
1558 
1559 	if (mask == x86_spec_ctrl_base)
1560 		return;
1561 
1562 	pr_info("Update user space SMT mitigation: STIBP %s\n",
1563 		mask & SPEC_CTRL_STIBP ? "always-on" : "off");
1564 	x86_spec_ctrl_base = mask;
1565 	on_each_cpu(update_stibp_msr, NULL, 1);
1566 }
1567 
1568 /* Update the static key controlling the evaluation of TIF_SPEC_IB */
update_indir_branch_cond(void)1569 static void update_indir_branch_cond(void)
1570 {
1571 	if (sched_smt_active())
1572 		static_branch_enable(&switch_to_cond_stibp);
1573 	else
1574 		static_branch_disable(&switch_to_cond_stibp);
1575 }
1576 
1577 #undef pr_fmt
1578 #define pr_fmt(fmt) fmt
1579 
1580 /* Update the static key controlling the MDS CPU buffer clear in idle */
update_mds_branch_idle(void)1581 static void update_mds_branch_idle(void)
1582 {
1583 	u64 ia32_cap = x86_read_arch_cap_msr();
1584 
1585 	/*
1586 	 * Enable the idle clearing if SMT is active on CPUs which are
1587 	 * affected only by MSBDS and not any other MDS variant.
1588 	 *
1589 	 * The other variants cannot be mitigated when SMT is enabled, so
1590 	 * clearing the buffers on idle just to prevent the Store Buffer
1591 	 * repartitioning leak would be a window dressing exercise.
1592 	 */
1593 	if (!boot_cpu_has_bug(X86_BUG_MSBDS_ONLY))
1594 		return;
1595 
1596 	if (sched_smt_active()) {
1597 		static_branch_enable(&mds_idle_clear);
1598 	} else if (mmio_mitigation == MMIO_MITIGATION_OFF ||
1599 		   (ia32_cap & ARCH_CAP_FBSDP_NO)) {
1600 		static_branch_disable(&mds_idle_clear);
1601 	}
1602 }
1603 
1604 #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"
1605 #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"
1606 #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"
1607 
cpu_bugs_smt_update(void)1608 void cpu_bugs_smt_update(void)
1609 {
1610 	mutex_lock(&spec_ctrl_mutex);
1611 
1612 	if (sched_smt_active() && unprivileged_ebpf_enabled() &&
1613 	    spectre_v2_enabled == SPECTRE_V2_EIBRS_LFENCE)
1614 		pr_warn_once(SPECTRE_V2_EIBRS_LFENCE_EBPF_SMT_MSG);
1615 
1616 	switch (spectre_v2_user_stibp) {
1617 	case SPECTRE_V2_USER_NONE:
1618 		break;
1619 	case SPECTRE_V2_USER_STRICT:
1620 	case SPECTRE_V2_USER_STRICT_PREFERRED:
1621 		update_stibp_strict();
1622 		break;
1623 	case SPECTRE_V2_USER_PRCTL:
1624 	case SPECTRE_V2_USER_SECCOMP:
1625 		update_indir_branch_cond();
1626 		break;
1627 	}
1628 
1629 	switch (mds_mitigation) {
1630 	case MDS_MITIGATION_FULL:
1631 	case MDS_MITIGATION_VMWERV:
1632 		if (sched_smt_active() && !boot_cpu_has(X86_BUG_MSBDS_ONLY))
1633 			pr_warn_once(MDS_MSG_SMT);
1634 		update_mds_branch_idle();
1635 		break;
1636 	case MDS_MITIGATION_OFF:
1637 		break;
1638 	}
1639 
1640 	switch (taa_mitigation) {
1641 	case TAA_MITIGATION_VERW:
1642 	case TAA_MITIGATION_UCODE_NEEDED:
1643 		if (sched_smt_active())
1644 			pr_warn_once(TAA_MSG_SMT);
1645 		break;
1646 	case TAA_MITIGATION_TSX_DISABLED:
1647 	case TAA_MITIGATION_OFF:
1648 		break;
1649 	}
1650 
1651 	switch (mmio_mitigation) {
1652 	case MMIO_MITIGATION_VERW:
1653 	case MMIO_MITIGATION_UCODE_NEEDED:
1654 		if (sched_smt_active())
1655 			pr_warn_once(MMIO_MSG_SMT);
1656 		break;
1657 	case MMIO_MITIGATION_OFF:
1658 		break;
1659 	}
1660 
1661 	mutex_unlock(&spec_ctrl_mutex);
1662 }
1663 
1664 #undef pr_fmt
1665 #define pr_fmt(fmt)	"Speculative Store Bypass: " fmt
1666 
1667 static enum ssb_mitigation ssb_mode __ro_after_init = SPEC_STORE_BYPASS_NONE;
1668 
1669 /* The kernel command line selection */
1670 enum ssb_mitigation_cmd {
1671 	SPEC_STORE_BYPASS_CMD_NONE,
1672 	SPEC_STORE_BYPASS_CMD_AUTO,
1673 	SPEC_STORE_BYPASS_CMD_ON,
1674 	SPEC_STORE_BYPASS_CMD_PRCTL,
1675 	SPEC_STORE_BYPASS_CMD_SECCOMP,
1676 };
1677 
1678 static const char * const ssb_strings[] = {
1679 	[SPEC_STORE_BYPASS_NONE]	= "Vulnerable",
1680 	[SPEC_STORE_BYPASS_DISABLE]	= "Mitigation: Speculative Store Bypass disabled",
1681 	[SPEC_STORE_BYPASS_PRCTL]	= "Mitigation: Speculative Store Bypass disabled via prctl",
1682 	[SPEC_STORE_BYPASS_SECCOMP]	= "Mitigation: Speculative Store Bypass disabled via prctl and seccomp",
1683 };
1684 
1685 static const struct {
1686 	const char *option;
1687 	enum ssb_mitigation_cmd cmd;
1688 } ssb_mitigation_options[]  __initconst = {
1689 	{ "auto",	SPEC_STORE_BYPASS_CMD_AUTO },    /* Platform decides */
1690 	{ "on",		SPEC_STORE_BYPASS_CMD_ON },      /* Disable Speculative Store Bypass */
1691 	{ "off",	SPEC_STORE_BYPASS_CMD_NONE },    /* Don't touch Speculative Store Bypass */
1692 	{ "prctl",	SPEC_STORE_BYPASS_CMD_PRCTL },   /* Disable Speculative Store Bypass via prctl */
1693 	{ "seccomp",	SPEC_STORE_BYPASS_CMD_SECCOMP }, /* Disable Speculative Store Bypass via prctl and seccomp */
1694 };
1695 
ssb_parse_cmdline(void)1696 static enum ssb_mitigation_cmd __init ssb_parse_cmdline(void)
1697 {
1698 	enum ssb_mitigation_cmd cmd = SPEC_STORE_BYPASS_CMD_AUTO;
1699 	char arg[20];
1700 	int ret, i;
1701 
1702 	if (cmdline_find_option_bool(boot_command_line, "nospec_store_bypass_disable") ||
1703 	    cpu_mitigations_off()) {
1704 		return SPEC_STORE_BYPASS_CMD_NONE;
1705 	} else {
1706 		ret = cmdline_find_option(boot_command_line, "spec_store_bypass_disable",
1707 					  arg, sizeof(arg));
1708 		if (ret < 0)
1709 			return SPEC_STORE_BYPASS_CMD_AUTO;
1710 
1711 		for (i = 0; i < ARRAY_SIZE(ssb_mitigation_options); i++) {
1712 			if (!match_option(arg, ret, ssb_mitigation_options[i].option))
1713 				continue;
1714 
1715 			cmd = ssb_mitigation_options[i].cmd;
1716 			break;
1717 		}
1718 
1719 		if (i >= ARRAY_SIZE(ssb_mitigation_options)) {
1720 			pr_err("unknown option (%s). Switching to AUTO select\n", arg);
1721 			return SPEC_STORE_BYPASS_CMD_AUTO;
1722 		}
1723 	}
1724 
1725 	return cmd;
1726 }
1727 
__ssb_select_mitigation(void)1728 static enum ssb_mitigation __init __ssb_select_mitigation(void)
1729 {
1730 	enum ssb_mitigation mode = SPEC_STORE_BYPASS_NONE;
1731 	enum ssb_mitigation_cmd cmd;
1732 
1733 	if (!boot_cpu_has(X86_FEATURE_SSBD))
1734 		return mode;
1735 
1736 	cmd = ssb_parse_cmdline();
1737 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS) &&
1738 	    (cmd == SPEC_STORE_BYPASS_CMD_NONE ||
1739 	     cmd == SPEC_STORE_BYPASS_CMD_AUTO))
1740 		return mode;
1741 
1742 	switch (cmd) {
1743 	case SPEC_STORE_BYPASS_CMD_AUTO:
1744 	case SPEC_STORE_BYPASS_CMD_SECCOMP:
1745 		/*
1746 		 * Choose prctl+seccomp as the default mode if seccomp is
1747 		 * enabled.
1748 		 */
1749 		if (IS_ENABLED(CONFIG_SECCOMP))
1750 			mode = SPEC_STORE_BYPASS_SECCOMP;
1751 		else
1752 			mode = SPEC_STORE_BYPASS_PRCTL;
1753 		break;
1754 	case SPEC_STORE_BYPASS_CMD_ON:
1755 		mode = SPEC_STORE_BYPASS_DISABLE;
1756 		break;
1757 	case SPEC_STORE_BYPASS_CMD_PRCTL:
1758 		mode = SPEC_STORE_BYPASS_PRCTL;
1759 		break;
1760 	case SPEC_STORE_BYPASS_CMD_NONE:
1761 		break;
1762 	}
1763 
1764 	/*
1765 	 * We have three CPU feature flags that are in play here:
1766 	 *  - X86_BUG_SPEC_STORE_BYPASS - CPU is susceptible.
1767 	 *  - X86_FEATURE_SSBD - CPU is able to turn off speculative store bypass
1768 	 *  - X86_FEATURE_SPEC_STORE_BYPASS_DISABLE - engage the mitigation
1769 	 */
1770 	if (mode == SPEC_STORE_BYPASS_DISABLE) {
1771 		setup_force_cpu_cap(X86_FEATURE_SPEC_STORE_BYPASS_DISABLE);
1772 		/*
1773 		 * Intel uses the SPEC CTRL MSR Bit(2) for this, while AMD may
1774 		 * use a completely different MSR and bit dependent on family.
1775 		 */
1776 		if (!static_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) &&
1777 		    !static_cpu_has(X86_FEATURE_AMD_SSBD)) {
1778 			x86_amd_ssb_disable();
1779 		} else {
1780 			x86_spec_ctrl_base |= SPEC_CTRL_SSBD;
1781 			update_spec_ctrl(x86_spec_ctrl_base);
1782 		}
1783 	}
1784 
1785 	return mode;
1786 }
1787 
ssb_select_mitigation(void)1788 static void ssb_select_mitigation(void)
1789 {
1790 	ssb_mode = __ssb_select_mitigation();
1791 
1792 	if (boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1793 		pr_info("%s\n", ssb_strings[ssb_mode]);
1794 }
1795 
1796 #undef pr_fmt
1797 #define pr_fmt(fmt)     "Speculation prctl: " fmt
1798 
task_update_spec_tif(struct task_struct * tsk)1799 static void task_update_spec_tif(struct task_struct *tsk)
1800 {
1801 	/* Force the update of the real TIF bits */
1802 	set_tsk_thread_flag(tsk, TIF_SPEC_FORCE_UPDATE);
1803 
1804 	/*
1805 	 * Immediately update the speculation control MSRs for the current
1806 	 * task, but for a non-current task delay setting the CPU
1807 	 * mitigation until it is scheduled next.
1808 	 *
1809 	 * This can only happen for SECCOMP mitigation. For PRCTL it's
1810 	 * always the current task.
1811 	 */
1812 	if (tsk == current)
1813 		speculation_ctrl_update_current();
1814 }
1815 
ssb_prctl_set(struct task_struct * task,unsigned long ctrl)1816 static int ssb_prctl_set(struct task_struct *task, unsigned long ctrl)
1817 {
1818 	if (ssb_mode != SPEC_STORE_BYPASS_PRCTL &&
1819 	    ssb_mode != SPEC_STORE_BYPASS_SECCOMP)
1820 		return -ENXIO;
1821 
1822 	switch (ctrl) {
1823 	case PR_SPEC_ENABLE:
1824 		/* If speculation is force disabled, enable is not allowed */
1825 		if (task_spec_ssb_force_disable(task))
1826 			return -EPERM;
1827 		task_clear_spec_ssb_disable(task);
1828 		task_clear_spec_ssb_noexec(task);
1829 		task_update_spec_tif(task);
1830 		break;
1831 	case PR_SPEC_DISABLE:
1832 		task_set_spec_ssb_disable(task);
1833 		task_clear_spec_ssb_noexec(task);
1834 		task_update_spec_tif(task);
1835 		break;
1836 	case PR_SPEC_FORCE_DISABLE:
1837 		task_set_spec_ssb_disable(task);
1838 		task_set_spec_ssb_force_disable(task);
1839 		task_clear_spec_ssb_noexec(task);
1840 		task_update_spec_tif(task);
1841 		break;
1842 	case PR_SPEC_DISABLE_NOEXEC:
1843 		if (task_spec_ssb_force_disable(task))
1844 			return -EPERM;
1845 		task_set_spec_ssb_disable(task);
1846 		task_set_spec_ssb_noexec(task);
1847 		task_update_spec_tif(task);
1848 		break;
1849 	default:
1850 		return -ERANGE;
1851 	}
1852 	return 0;
1853 }
1854 
is_spec_ib_user_controlled(void)1855 static bool is_spec_ib_user_controlled(void)
1856 {
1857 	return spectre_v2_user_ibpb == SPECTRE_V2_USER_PRCTL ||
1858 		spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP ||
1859 		spectre_v2_user_stibp == SPECTRE_V2_USER_PRCTL ||
1860 		spectre_v2_user_stibp == SPECTRE_V2_USER_SECCOMP;
1861 }
1862 
ib_prctl_set(struct task_struct * task,unsigned long ctrl)1863 static int ib_prctl_set(struct task_struct *task, unsigned long ctrl)
1864 {
1865 	switch (ctrl) {
1866 	case PR_SPEC_ENABLE:
1867 		if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
1868 		    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
1869 			return 0;
1870 
1871 		/*
1872 		 * With strict mode for both IBPB and STIBP, the instruction
1873 		 * code paths avoid checking this task flag and instead,
1874 		 * unconditionally run the instruction. However, STIBP and IBPB
1875 		 * are independent and either can be set to conditionally
1876 		 * enabled regardless of the mode of the other.
1877 		 *
1878 		 * If either is set to conditional, allow the task flag to be
1879 		 * updated, unless it was force-disabled by a previous prctl
1880 		 * call. Currently, this is possible on an AMD CPU which has the
1881 		 * feature X86_FEATURE_AMD_STIBP_ALWAYS_ON. In this case, if the
1882 		 * kernel is booted with 'spectre_v2_user=seccomp', then
1883 		 * spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP and
1884 		 * spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED.
1885 		 */
1886 		if (!is_spec_ib_user_controlled() ||
1887 		    task_spec_ib_force_disable(task))
1888 			return -EPERM;
1889 
1890 		task_clear_spec_ib_disable(task);
1891 		task_update_spec_tif(task);
1892 		break;
1893 	case PR_SPEC_DISABLE:
1894 	case PR_SPEC_FORCE_DISABLE:
1895 		/*
1896 		 * Indirect branch speculation is always allowed when
1897 		 * mitigation is force disabled.
1898 		 */
1899 		if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
1900 		    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
1901 			return -EPERM;
1902 
1903 		if (!is_spec_ib_user_controlled())
1904 			return 0;
1905 
1906 		task_set_spec_ib_disable(task);
1907 		if (ctrl == PR_SPEC_FORCE_DISABLE)
1908 			task_set_spec_ib_force_disable(task);
1909 		task_update_spec_tif(task);
1910 		if (task == current)
1911 			indirect_branch_prediction_barrier();
1912 		break;
1913 	default:
1914 		return -ERANGE;
1915 	}
1916 	return 0;
1917 }
1918 
arch_prctl_spec_ctrl_set(struct task_struct * task,unsigned long which,unsigned long ctrl)1919 int arch_prctl_spec_ctrl_set(struct task_struct *task, unsigned long which,
1920 			     unsigned long ctrl)
1921 {
1922 	switch (which) {
1923 	case PR_SPEC_STORE_BYPASS:
1924 		return ssb_prctl_set(task, ctrl);
1925 	case PR_SPEC_INDIRECT_BRANCH:
1926 		return ib_prctl_set(task, ctrl);
1927 	default:
1928 		return -ENODEV;
1929 	}
1930 }
1931 
1932 #ifdef CONFIG_SECCOMP
arch_seccomp_spec_mitigate(struct task_struct * task)1933 void arch_seccomp_spec_mitigate(struct task_struct *task)
1934 {
1935 	if (ssb_mode == SPEC_STORE_BYPASS_SECCOMP)
1936 		ssb_prctl_set(task, PR_SPEC_FORCE_DISABLE);
1937 	if (spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP ||
1938 	    spectre_v2_user_stibp == SPECTRE_V2_USER_SECCOMP)
1939 		ib_prctl_set(task, PR_SPEC_FORCE_DISABLE);
1940 }
1941 #endif
1942 
ssb_prctl_get(struct task_struct * task)1943 static int ssb_prctl_get(struct task_struct *task)
1944 {
1945 	switch (ssb_mode) {
1946 	case SPEC_STORE_BYPASS_DISABLE:
1947 		return PR_SPEC_DISABLE;
1948 	case SPEC_STORE_BYPASS_SECCOMP:
1949 	case SPEC_STORE_BYPASS_PRCTL:
1950 		if (task_spec_ssb_force_disable(task))
1951 			return PR_SPEC_PRCTL | PR_SPEC_FORCE_DISABLE;
1952 		if (task_spec_ssb_noexec(task))
1953 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE_NOEXEC;
1954 		if (task_spec_ssb_disable(task))
1955 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
1956 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
1957 	default:
1958 		if (boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1959 			return PR_SPEC_ENABLE;
1960 		return PR_SPEC_NOT_AFFECTED;
1961 	}
1962 }
1963 
ib_prctl_get(struct task_struct * task)1964 static int ib_prctl_get(struct task_struct *task)
1965 {
1966 	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V2))
1967 		return PR_SPEC_NOT_AFFECTED;
1968 
1969 	if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
1970 	    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
1971 		return PR_SPEC_ENABLE;
1972 	else if (is_spec_ib_user_controlled()) {
1973 		if (task_spec_ib_force_disable(task))
1974 			return PR_SPEC_PRCTL | PR_SPEC_FORCE_DISABLE;
1975 		if (task_spec_ib_disable(task))
1976 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
1977 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
1978 	} else if (spectre_v2_user_ibpb == SPECTRE_V2_USER_STRICT ||
1979 	    spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT ||
1980 	    spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED)
1981 		return PR_SPEC_DISABLE;
1982 	else
1983 		return PR_SPEC_NOT_AFFECTED;
1984 }
1985 
arch_prctl_spec_ctrl_get(struct task_struct * task,unsigned long which)1986 int arch_prctl_spec_ctrl_get(struct task_struct *task, unsigned long which)
1987 {
1988 	switch (which) {
1989 	case PR_SPEC_STORE_BYPASS:
1990 		return ssb_prctl_get(task);
1991 	case PR_SPEC_INDIRECT_BRANCH:
1992 		return ib_prctl_get(task);
1993 	default:
1994 		return -ENODEV;
1995 	}
1996 }
1997 
x86_spec_ctrl_setup_ap(void)1998 void x86_spec_ctrl_setup_ap(void)
1999 {
2000 	if (boot_cpu_has(X86_FEATURE_MSR_SPEC_CTRL))
2001 		update_spec_ctrl(x86_spec_ctrl_base);
2002 
2003 	if (ssb_mode == SPEC_STORE_BYPASS_DISABLE)
2004 		x86_amd_ssb_disable();
2005 }
2006 
2007 bool itlb_multihit_kvm_mitigation;
2008 EXPORT_SYMBOL_GPL(itlb_multihit_kvm_mitigation);
2009 
2010 #undef pr_fmt
2011 #define pr_fmt(fmt)	"L1TF: " fmt
2012 
2013 /* Default mitigation for L1TF-affected CPUs */
2014 enum l1tf_mitigations l1tf_mitigation __ro_after_init = L1TF_MITIGATION_FLUSH;
2015 #if IS_ENABLED(CONFIG_KVM_INTEL)
2016 EXPORT_SYMBOL_GPL(l1tf_mitigation);
2017 #endif
2018 enum vmx_l1d_flush_state l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
2019 EXPORT_SYMBOL_GPL(l1tf_vmx_mitigation);
2020 
2021 /*
2022  * These CPUs all support 44bits physical address space internally in the
2023  * cache but CPUID can report a smaller number of physical address bits.
2024  *
2025  * The L1TF mitigation uses the top most address bit for the inversion of
2026  * non present PTEs. When the installed memory reaches into the top most
2027  * address bit due to memory holes, which has been observed on machines
2028  * which report 36bits physical address bits and have 32G RAM installed,
2029  * then the mitigation range check in l1tf_select_mitigation() triggers.
2030  * This is a false positive because the mitigation is still possible due to
2031  * the fact that the cache uses 44bit internally. Use the cache bits
2032  * instead of the reported physical bits and adjust them on the affected
2033  * machines to 44bit if the reported bits are less than 44.
2034  */
override_cache_bits(struct cpuinfo_x86 * c)2035 static void override_cache_bits(struct cpuinfo_x86 *c)
2036 {
2037 	if (c->x86 != 6)
2038 		return;
2039 
2040 	switch (c->x86_model) {
2041 	case INTEL_FAM6_NEHALEM:
2042 	case INTEL_FAM6_WESTMERE:
2043 	case INTEL_FAM6_SANDYBRIDGE:
2044 	case INTEL_FAM6_IVYBRIDGE:
2045 	case INTEL_FAM6_HASWELL:
2046 	case INTEL_FAM6_HASWELL_L:
2047 	case INTEL_FAM6_HASWELL_G:
2048 	case INTEL_FAM6_BROADWELL:
2049 	case INTEL_FAM6_BROADWELL_G:
2050 	case INTEL_FAM6_SKYLAKE_L:
2051 	case INTEL_FAM6_SKYLAKE:
2052 	case INTEL_FAM6_KABYLAKE_L:
2053 	case INTEL_FAM6_KABYLAKE:
2054 		if (c->x86_cache_bits < 44)
2055 			c->x86_cache_bits = 44;
2056 		break;
2057 	}
2058 }
2059 
l1tf_select_mitigation(void)2060 static void __init l1tf_select_mitigation(void)
2061 {
2062 	u64 half_pa;
2063 
2064 	if (!boot_cpu_has_bug(X86_BUG_L1TF))
2065 		return;
2066 
2067 	if (cpu_mitigations_off())
2068 		l1tf_mitigation = L1TF_MITIGATION_OFF;
2069 	else if (cpu_mitigations_auto_nosmt())
2070 		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOSMT;
2071 
2072 	override_cache_bits(&boot_cpu_data);
2073 
2074 	switch (l1tf_mitigation) {
2075 	case L1TF_MITIGATION_OFF:
2076 	case L1TF_MITIGATION_FLUSH_NOWARN:
2077 	case L1TF_MITIGATION_FLUSH:
2078 		break;
2079 	case L1TF_MITIGATION_FLUSH_NOSMT:
2080 	case L1TF_MITIGATION_FULL:
2081 		cpu_smt_disable(false);
2082 		break;
2083 	case L1TF_MITIGATION_FULL_FORCE:
2084 		cpu_smt_disable(true);
2085 		break;
2086 	}
2087 
2088 #if CONFIG_PGTABLE_LEVELS == 2
2089 	pr_warn("Kernel not compiled for PAE. No mitigation for L1TF\n");
2090 	return;
2091 #endif
2092 
2093 	half_pa = (u64)l1tf_pfn_limit() << PAGE_SHIFT;
2094 	if (l1tf_mitigation != L1TF_MITIGATION_OFF &&
2095 			e820__mapped_any(half_pa, ULLONG_MAX - half_pa, E820_TYPE_RAM)) {
2096 		pr_warn("System has more than MAX_PA/2 memory. L1TF mitigation not effective.\n");
2097 		pr_info("You may make it effective by booting the kernel with mem=%llu parameter.\n",
2098 				half_pa);
2099 		pr_info("However, doing so will make a part of your RAM unusable.\n");
2100 		pr_info("Reading https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/l1tf.html might help you decide.\n");
2101 		return;
2102 	}
2103 
2104 	setup_force_cpu_cap(X86_FEATURE_L1TF_PTEINV);
2105 }
2106 
l1tf_cmdline(char * str)2107 static int __init l1tf_cmdline(char *str)
2108 {
2109 	if (!boot_cpu_has_bug(X86_BUG_L1TF))
2110 		return 0;
2111 
2112 	if (!str)
2113 		return -EINVAL;
2114 
2115 	if (!strcmp(str, "off"))
2116 		l1tf_mitigation = L1TF_MITIGATION_OFF;
2117 	else if (!strcmp(str, "flush,nowarn"))
2118 		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOWARN;
2119 	else if (!strcmp(str, "flush"))
2120 		l1tf_mitigation = L1TF_MITIGATION_FLUSH;
2121 	else if (!strcmp(str, "flush,nosmt"))
2122 		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOSMT;
2123 	else if (!strcmp(str, "full"))
2124 		l1tf_mitigation = L1TF_MITIGATION_FULL;
2125 	else if (!strcmp(str, "full,force"))
2126 		l1tf_mitigation = L1TF_MITIGATION_FULL_FORCE;
2127 
2128 	return 0;
2129 }
2130 early_param("l1tf", l1tf_cmdline);
2131 
2132 #undef pr_fmt
2133 #define pr_fmt(fmt) fmt
2134 
2135 #ifdef CONFIG_SYSFS
2136 
2137 #define L1TF_DEFAULT_MSG "Mitigation: PTE Inversion"
2138 
2139 #if IS_ENABLED(CONFIG_KVM_INTEL)
2140 static const char * const l1tf_vmx_states[] = {
2141 	[VMENTER_L1D_FLUSH_AUTO]		= "auto",
2142 	[VMENTER_L1D_FLUSH_NEVER]		= "vulnerable",
2143 	[VMENTER_L1D_FLUSH_COND]		= "conditional cache flushes",
2144 	[VMENTER_L1D_FLUSH_ALWAYS]		= "cache flushes",
2145 	[VMENTER_L1D_FLUSH_EPT_DISABLED]	= "EPT disabled",
2146 	[VMENTER_L1D_FLUSH_NOT_REQUIRED]	= "flush not necessary"
2147 };
2148 
l1tf_show_state(char * buf)2149 static ssize_t l1tf_show_state(char *buf)
2150 {
2151 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO)
2152 		return sprintf(buf, "%s\n", L1TF_DEFAULT_MSG);
2153 
2154 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_EPT_DISABLED ||
2155 	    (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER &&
2156 	     sched_smt_active())) {
2157 		return sprintf(buf, "%s; VMX: %s\n", L1TF_DEFAULT_MSG,
2158 			       l1tf_vmx_states[l1tf_vmx_mitigation]);
2159 	}
2160 
2161 	return sprintf(buf, "%s; VMX: %s, SMT %s\n", L1TF_DEFAULT_MSG,
2162 		       l1tf_vmx_states[l1tf_vmx_mitigation],
2163 		       sched_smt_active() ? "vulnerable" : "disabled");
2164 }
2165 
itlb_multihit_show_state(char * buf)2166 static ssize_t itlb_multihit_show_state(char *buf)
2167 {
2168 	if (!boot_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
2169 	    !boot_cpu_has(X86_FEATURE_VMX))
2170 		return sprintf(buf, "KVM: Mitigation: VMX unsupported\n");
2171 	else if (!(cr4_read_shadow() & X86_CR4_VMXE))
2172 		return sprintf(buf, "KVM: Mitigation: VMX disabled\n");
2173 	else if (itlb_multihit_kvm_mitigation)
2174 		return sprintf(buf, "KVM: Mitigation: Split huge pages\n");
2175 	else
2176 		return sprintf(buf, "KVM: Vulnerable\n");
2177 }
2178 #else
l1tf_show_state(char * buf)2179 static ssize_t l1tf_show_state(char *buf)
2180 {
2181 	return sprintf(buf, "%s\n", L1TF_DEFAULT_MSG);
2182 }
2183 
itlb_multihit_show_state(char * buf)2184 static ssize_t itlb_multihit_show_state(char *buf)
2185 {
2186 	return sprintf(buf, "Processor vulnerable\n");
2187 }
2188 #endif
2189 
mds_show_state(char * buf)2190 static ssize_t mds_show_state(char *buf)
2191 {
2192 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2193 		return sprintf(buf, "%s; SMT Host state unknown\n",
2194 			       mds_strings[mds_mitigation]);
2195 	}
2196 
2197 	if (boot_cpu_has(X86_BUG_MSBDS_ONLY)) {
2198 		return sprintf(buf, "%s; SMT %s\n", mds_strings[mds_mitigation],
2199 			       (mds_mitigation == MDS_MITIGATION_OFF ? "vulnerable" :
2200 			        sched_smt_active() ? "mitigated" : "disabled"));
2201 	}
2202 
2203 	return sprintf(buf, "%s; SMT %s\n", mds_strings[mds_mitigation],
2204 		       sched_smt_active() ? "vulnerable" : "disabled");
2205 }
2206 
tsx_async_abort_show_state(char * buf)2207 static ssize_t tsx_async_abort_show_state(char *buf)
2208 {
2209 	if ((taa_mitigation == TAA_MITIGATION_TSX_DISABLED) ||
2210 	    (taa_mitigation == TAA_MITIGATION_OFF))
2211 		return sprintf(buf, "%s\n", taa_strings[taa_mitigation]);
2212 
2213 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2214 		return sprintf(buf, "%s; SMT Host state unknown\n",
2215 			       taa_strings[taa_mitigation]);
2216 	}
2217 
2218 	return sprintf(buf, "%s; SMT %s\n", taa_strings[taa_mitigation],
2219 		       sched_smt_active() ? "vulnerable" : "disabled");
2220 }
2221 
mmio_stale_data_show_state(char * buf)2222 static ssize_t mmio_stale_data_show_state(char *buf)
2223 {
2224 	if (boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN))
2225 		return sysfs_emit(buf, "Unknown: No mitigations\n");
2226 
2227 	if (mmio_mitigation == MMIO_MITIGATION_OFF)
2228 		return sysfs_emit(buf, "%s\n", mmio_strings[mmio_mitigation]);
2229 
2230 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2231 		return sysfs_emit(buf, "%s; SMT Host state unknown\n",
2232 				  mmio_strings[mmio_mitigation]);
2233 	}
2234 
2235 	return sysfs_emit(buf, "%s; SMT %s\n", mmio_strings[mmio_mitigation],
2236 			  sched_smt_active() ? "vulnerable" : "disabled");
2237 }
2238 
stibp_state(void)2239 static char *stibp_state(void)
2240 {
2241 	if (spectre_v2_in_eibrs_mode(spectre_v2_enabled))
2242 		return "";
2243 
2244 	switch (spectre_v2_user_stibp) {
2245 	case SPECTRE_V2_USER_NONE:
2246 		return ", STIBP: disabled";
2247 	case SPECTRE_V2_USER_STRICT:
2248 		return ", STIBP: forced";
2249 	case SPECTRE_V2_USER_STRICT_PREFERRED:
2250 		return ", STIBP: always-on";
2251 	case SPECTRE_V2_USER_PRCTL:
2252 	case SPECTRE_V2_USER_SECCOMP:
2253 		if (static_key_enabled(&switch_to_cond_stibp))
2254 			return ", STIBP: conditional";
2255 	}
2256 	return "";
2257 }
2258 
ibpb_state(void)2259 static char *ibpb_state(void)
2260 {
2261 	if (boot_cpu_has(X86_FEATURE_IBPB)) {
2262 		if (static_key_enabled(&switch_mm_always_ibpb))
2263 			return ", IBPB: always-on";
2264 		if (static_key_enabled(&switch_mm_cond_ibpb))
2265 			return ", IBPB: conditional";
2266 		return ", IBPB: disabled";
2267 	}
2268 	return "";
2269 }
2270 
pbrsb_eibrs_state(void)2271 static char *pbrsb_eibrs_state(void)
2272 {
2273 	if (boot_cpu_has_bug(X86_BUG_EIBRS_PBRSB)) {
2274 		if (boot_cpu_has(X86_FEATURE_RSB_VMEXIT_LITE) ||
2275 		    boot_cpu_has(X86_FEATURE_RSB_VMEXIT))
2276 			return ", PBRSB-eIBRS: SW sequence";
2277 		else
2278 			return ", PBRSB-eIBRS: Vulnerable";
2279 	} else {
2280 		return ", PBRSB-eIBRS: Not affected";
2281 	}
2282 }
2283 
spectre_v2_show_state(char * buf)2284 static ssize_t spectre_v2_show_state(char *buf)
2285 {
2286 	if (spectre_v2_enabled == SPECTRE_V2_LFENCE)
2287 		return sprintf(buf, "Vulnerable: LFENCE\n");
2288 
2289 	if (spectre_v2_enabled == SPECTRE_V2_EIBRS && unprivileged_ebpf_enabled())
2290 		return sprintf(buf, "Vulnerable: eIBRS with unprivileged eBPF\n");
2291 
2292 	if (sched_smt_active() && unprivileged_ebpf_enabled() &&
2293 	    spectre_v2_enabled == SPECTRE_V2_EIBRS_LFENCE)
2294 		return sprintf(buf, "Vulnerable: eIBRS+LFENCE with unprivileged eBPF and SMT\n");
2295 
2296 	return sprintf(buf, "%s%s%s%s%s%s%s\n",
2297 		       spectre_v2_strings[spectre_v2_enabled],
2298 		       ibpb_state(),
2299 		       boot_cpu_has(X86_FEATURE_USE_IBRS_FW) ? ", IBRS_FW" : "",
2300 		       stibp_state(),
2301 		       boot_cpu_has(X86_FEATURE_RSB_CTXSW) ? ", RSB filling" : "",
2302 		       pbrsb_eibrs_state(),
2303 		       spectre_v2_module_string());
2304 }
2305 
srbds_show_state(char * buf)2306 static ssize_t srbds_show_state(char *buf)
2307 {
2308 	return sprintf(buf, "%s\n", srbds_strings[srbds_mitigation]);
2309 }
2310 
retbleed_show_state(char * buf)2311 static ssize_t retbleed_show_state(char *buf)
2312 {
2313 	if (retbleed_mitigation == RETBLEED_MITIGATION_UNRET ||
2314 	    retbleed_mitigation == RETBLEED_MITIGATION_IBPB) {
2315 	    if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD &&
2316 		boot_cpu_data.x86_vendor != X86_VENDOR_HYGON)
2317 		    return sprintf(buf, "Vulnerable: untrained return thunk / IBPB on non-AMD based uarch\n");
2318 
2319 	    return sprintf(buf, "%s; SMT %s\n",
2320 			   retbleed_strings[retbleed_mitigation],
2321 			   !sched_smt_active() ? "disabled" :
2322 			   spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT ||
2323 			   spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED ?
2324 			   "enabled with STIBP protection" : "vulnerable");
2325 	}
2326 
2327 	return sprintf(buf, "%s\n", retbleed_strings[retbleed_mitigation]);
2328 }
2329 
cpu_show_common(struct device * dev,struct device_attribute * attr,char * buf,unsigned int bug)2330 static ssize_t cpu_show_common(struct device *dev, struct device_attribute *attr,
2331 			       char *buf, unsigned int bug)
2332 {
2333 	if (!boot_cpu_has_bug(bug))
2334 		return sprintf(buf, "Not affected\n");
2335 
2336 	switch (bug) {
2337 	case X86_BUG_CPU_MELTDOWN:
2338 		if (boot_cpu_has(X86_FEATURE_PTI))
2339 			return sprintf(buf, "Mitigation: PTI\n");
2340 
2341 		if (hypervisor_is_type(X86_HYPER_XEN_PV))
2342 			return sprintf(buf, "Unknown (XEN PV detected, hypervisor mitigation required)\n");
2343 
2344 		break;
2345 
2346 	case X86_BUG_SPECTRE_V1:
2347 		return sprintf(buf, "%s\n", spectre_v1_strings[spectre_v1_mitigation]);
2348 
2349 	case X86_BUG_SPECTRE_V2:
2350 		return spectre_v2_show_state(buf);
2351 
2352 	case X86_BUG_SPEC_STORE_BYPASS:
2353 		return sprintf(buf, "%s\n", ssb_strings[ssb_mode]);
2354 
2355 	case X86_BUG_L1TF:
2356 		if (boot_cpu_has(X86_FEATURE_L1TF_PTEINV))
2357 			return l1tf_show_state(buf);
2358 		break;
2359 
2360 	case X86_BUG_MDS:
2361 		return mds_show_state(buf);
2362 
2363 	case X86_BUG_TAA:
2364 		return tsx_async_abort_show_state(buf);
2365 
2366 	case X86_BUG_ITLB_MULTIHIT:
2367 		return itlb_multihit_show_state(buf);
2368 
2369 	case X86_BUG_SRBDS:
2370 		return srbds_show_state(buf);
2371 
2372 	case X86_BUG_MMIO_STALE_DATA:
2373 	case X86_BUG_MMIO_UNKNOWN:
2374 		return mmio_stale_data_show_state(buf);
2375 
2376 	case X86_BUG_RETBLEED:
2377 		return retbleed_show_state(buf);
2378 
2379 	default:
2380 		break;
2381 	}
2382 
2383 	return sprintf(buf, "Vulnerable\n");
2384 }
2385 
cpu_show_meltdown(struct device * dev,struct device_attribute * attr,char * buf)2386 ssize_t cpu_show_meltdown(struct device *dev, struct device_attribute *attr, char *buf)
2387 {
2388 	return cpu_show_common(dev, attr, buf, X86_BUG_CPU_MELTDOWN);
2389 }
2390 
cpu_show_spectre_v1(struct device * dev,struct device_attribute * attr,char * buf)2391 ssize_t cpu_show_spectre_v1(struct device *dev, struct device_attribute *attr, char *buf)
2392 {
2393 	return cpu_show_common(dev, attr, buf, X86_BUG_SPECTRE_V1);
2394 }
2395 
cpu_show_spectre_v2(struct device * dev,struct device_attribute * attr,char * buf)2396 ssize_t cpu_show_spectre_v2(struct device *dev, struct device_attribute *attr, char *buf)
2397 {
2398 	return cpu_show_common(dev, attr, buf, X86_BUG_SPECTRE_V2);
2399 }
2400 
cpu_show_spec_store_bypass(struct device * dev,struct device_attribute * attr,char * buf)2401 ssize_t cpu_show_spec_store_bypass(struct device *dev, struct device_attribute *attr, char *buf)
2402 {
2403 	return cpu_show_common(dev, attr, buf, X86_BUG_SPEC_STORE_BYPASS);
2404 }
2405 
cpu_show_l1tf(struct device * dev,struct device_attribute * attr,char * buf)2406 ssize_t cpu_show_l1tf(struct device *dev, struct device_attribute *attr, char *buf)
2407 {
2408 	return cpu_show_common(dev, attr, buf, X86_BUG_L1TF);
2409 }
2410 
cpu_show_mds(struct device * dev,struct device_attribute * attr,char * buf)2411 ssize_t cpu_show_mds(struct device *dev, struct device_attribute *attr, char *buf)
2412 {
2413 	return cpu_show_common(dev, attr, buf, X86_BUG_MDS);
2414 }
2415 
cpu_show_tsx_async_abort(struct device * dev,struct device_attribute * attr,char * buf)2416 ssize_t cpu_show_tsx_async_abort(struct device *dev, struct device_attribute *attr, char *buf)
2417 {
2418 	return cpu_show_common(dev, attr, buf, X86_BUG_TAA);
2419 }
2420 
cpu_show_itlb_multihit(struct device * dev,struct device_attribute * attr,char * buf)2421 ssize_t cpu_show_itlb_multihit(struct device *dev, struct device_attribute *attr, char *buf)
2422 {
2423 	return cpu_show_common(dev, attr, buf, X86_BUG_ITLB_MULTIHIT);
2424 }
2425 
cpu_show_srbds(struct device * dev,struct device_attribute * attr,char * buf)2426 ssize_t cpu_show_srbds(struct device *dev, struct device_attribute *attr, char *buf)
2427 {
2428 	return cpu_show_common(dev, attr, buf, X86_BUG_SRBDS);
2429 }
2430 
cpu_show_mmio_stale_data(struct device * dev,struct device_attribute * attr,char * buf)2431 ssize_t cpu_show_mmio_stale_data(struct device *dev, struct device_attribute *attr, char *buf)
2432 {
2433 	if (boot_cpu_has_bug(X86_BUG_MMIO_UNKNOWN))
2434 		return cpu_show_common(dev, attr, buf, X86_BUG_MMIO_UNKNOWN);
2435 	else
2436 		return cpu_show_common(dev, attr, buf, X86_BUG_MMIO_STALE_DATA);
2437 }
2438 
cpu_show_retbleed(struct device * dev,struct device_attribute * attr,char * buf)2439 ssize_t cpu_show_retbleed(struct device *dev, struct device_attribute *attr, char *buf)
2440 {
2441 	return cpu_show_common(dev, attr, buf, X86_BUG_RETBLEED);
2442 }
2443 #endif
2444