Linux Audio

Check our new training course

Loading...
v3.5.6
 
  1/*
  2 *  Copyright (C) 1994  Linus Torvalds
  3 *
  4 *  Cyrix stuff, June 1998 by:
  5 *	- Rafael R. Reilova (moved everything from head.S),
  6 *        <rreilova@ececs.uc.edu>
  7 *	- Channing Corn (tests & fixes),
  8 *	- Andrew D. Balsa (code cleanup).
  9 */
 10#include <linux/init.h>
 11#include <linux/utsname.h>
 
 
 
 
 
 
 
 
 
 12#include <asm/bugs.h>
 13#include <asm/processor.h>
 14#include <asm/processor-flags.h>
 15#include <asm/i387.h>
 16#include <asm/msr.h>
 
 17#include <asm/paravirt.h>
 18#include <asm/alternative.h>
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 19
 20static int __init no_halt(char *s)
 21{
 22	WARN_ONCE(1, "\"no-hlt\" is deprecated, please use \"idle=poll\"\n");
 23	boot_cpu_data.hlt_works_ok = 0;
 24	return 1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 25}
 26
 27__setup("no-hlt", no_halt);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 28
 29static int __init no_387(char *s)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 30{
 31	boot_cpu_data.hard_math = 0;
 32	write_cr0(X86_CR0_TS | X86_CR0_EM | X86_CR0_MP | read_cr0());
 33	return 1;
 
 
 
 34}
 35
 36__setup("no387", no_387);
 
 37
 38static double __initdata x = 4195835.0;
 39static double __initdata y = 3145727.0;
 
 40
 41/*
 42 * This used to check for exceptions..
 43 * However, it turns out that to support that,
 44 * the XMM trap handlers basically had to
 45 * be buggy. So let's have a correct XMM trap
 46 * handler, and forget about printing out
 47 * some status at boot.
 48 *
 49 * We should really only care about bugs here
 50 * anyway. Not features.
 51 */
 52static void __init check_fpu(void)
 53{
 54	s32 fdiv_bug;
 
 
 
 55
 56	if (!boot_cpu_data.hard_math) {
 57#ifndef CONFIG_MATH_EMULATION
 58		printk(KERN_EMERG "No coprocessor found and no math emulation present.\n");
 59		printk(KERN_EMERG "Giving up.\n");
 60		for (;;) ;
 61#endif
 
 
 
 
 
 
 
 
 
 62		return;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 63	}
 64
 65	kernel_fpu_begin();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 66
 67	/*
 68	 * trap_init() enabled FXSR and company _before_ testing for FP
 69	 * problems here.
 70	 *
 71	 * Test for the divl bug..
 
 72	 */
 73	__asm__("fninit\n\t"
 74		"fldl %1\n\t"
 75		"fdivl %2\n\t"
 76		"fmull %2\n\t"
 77		"fldl %1\n\t"
 78		"fsubp %%st,%%st(1)\n\t"
 79		"fistpl %0\n\t"
 80		"fwait\n\t"
 81		"fninit"
 82		: "=m" (*&fdiv_bug)
 83		: "m" (*&x), "m" (*&y));
 84
 85	kernel_fpu_end();
 
 86
 87	boot_cpu_data.fdiv_bug = fdiv_bug;
 88	if (boot_cpu_data.fdiv_bug)
 89		printk(KERN_WARNING "Hmm, FPU with FDIV bug.\n");
 
 
 
 
 
 
 
 
 90}
 91
 92static void __init check_hlt(void)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 93{
 94	if (boot_cpu_data.x86 >= 5 || paravirt_enabled())
 
 
 
 
 
 95		return;
 96
 97	printk(KERN_INFO "Checking 'hlt' instruction... ");
 98	if (!boot_cpu_data.hlt_works_ok) {
 99		printk("disabled\n");
100		return;
 
 
 
 
 
 
 
 
 
 
 
 
 
101	}
102	halt();
103	halt();
104	halt();
105	halt();
106	printk(KERN_CONT "OK.\n");
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
107}
108
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
109/*
110 *	Most 386 processors have a bug where a POPAD can lock the
111 *	machine even from user space.
112 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
113
114static void __init check_popad(void)
115{
116#ifndef CONFIG_X86_POPAD_OK
117	int res, inp = (int) &res;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
118
119	printk(KERN_INFO "Checking for popad bug... ");
120	__asm__ __volatile__(
121	  "movl $12345678,%%eax; movl $0,%%edi; pusha; popa; movl (%%edx,%%edi),%%ecx "
122	  : "=&a" (res)
123	  : "d" (inp)
124	  : "ecx", "edi");
125	/*
126	 * If this fails, it means that any user program may lock the
127	 * CPU hard. Too bad.
128	 */
129	if (res != 12345678)
130		printk(KERN_CONT "Buggy.\n");
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
131	else
132		printk(KERN_CONT "OK.\n");
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
133#endif
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
134}
135
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
136/*
137 * Check whether we are able to run this kernel safely on SMP.
 
138 *
139 * - In order to run on a i386, we need to be compiled for i386
140 *   (for due to lack of "invlpg" and working WP on a i386)
141 * - In order to run on anything without a TSC, we need to be
142 *   compiled for a i486.
 
 
 
 
 
143 */
 
 
 
 
144
145static void __init check_config(void)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
146{
147/*
148 * We'd better not be a i386 if we're configured to use some
149 * i486+ only features! (WP works in supervisor mode and the
150 * new "invlpg" and "bswap" instructions)
151 */
152#if defined(CONFIG_X86_WP_WORKS_OK) || defined(CONFIG_X86_INVLPG) || \
153	defined(CONFIG_X86_BSWAP)
154	if (boot_cpu_data.x86 == 3)
155		panic("Kernel requires i486+ for 'invlpg' and other features");
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
156#endif
 
 
 
 
 
 
 
 
 
 
 
 
 
157}
158
 
 
 
 
 
 
 
159
160void __init check_bugs(void)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
161{
162	identify_boot_cpu();
163#ifndef CONFIG_SMP
164	printk(KERN_INFO "CPU: ");
165	print_cpu_info(&boot_cpu_data);
166#endif
167	check_config();
168	check_fpu();
169	check_hlt();
170	check_popad();
171	init_utsname()->machine[1] =
172		'0' + (boot_cpu_data.x86 > 6 ? 6 : boot_cpu_data.x86);
173	alternative_instructions();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
174}
v5.14.15
   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
  20#include <asm/spec-ctrl.h>
  21#include <asm/cmdline.h>
  22#include <asm/bugs.h>
  23#include <asm/processor.h>
  24#include <asm/processor-flags.h>
  25#include <asm/fpu/internal.h>
  26#include <asm/msr.h>
  27#include <asm/vmx.h>
  28#include <asm/paravirt.h>
  29#include <asm/alternative.h>
  30#include <asm/set_memory.h>
  31#include <asm/intel-family.h>
  32#include <asm/e820/api.h>
  33#include <asm/hypervisor.h>
  34#include <asm/tlbflush.h>
  35
  36#include "cpu.h"
  37
  38static void __init spectre_v1_select_mitigation(void);
  39static void __init spectre_v2_select_mitigation(void);
  40static void __init ssb_select_mitigation(void);
  41static void __init l1tf_select_mitigation(void);
  42static void __init mds_select_mitigation(void);
  43static void __init mds_print_mitigation(void);
  44static void __init taa_select_mitigation(void);
  45static void __init srbds_select_mitigation(void);
  46
  47/* The base value of the SPEC_CTRL MSR that always has to be preserved. */
  48u64 x86_spec_ctrl_base;
  49EXPORT_SYMBOL_GPL(x86_spec_ctrl_base);
  50static DEFINE_MUTEX(spec_ctrl_mutex);
  51
  52/*
  53 * The vendor and possibly platform specific bits which can be modified in
  54 * x86_spec_ctrl_base.
  55 */
  56static u64 __ro_after_init x86_spec_ctrl_mask = SPEC_CTRL_IBRS;
  57
  58/*
  59 * AMD specific MSR info for Speculative Store Bypass control.
  60 * x86_amd_ls_cfg_ssbd_mask is initialized in identify_boot_cpu().
  61 */
  62u64 __ro_after_init x86_amd_ls_cfg_base;
  63u64 __ro_after_init x86_amd_ls_cfg_ssbd_mask;
  64
  65/* Control conditional STIBP in switch_to() */
  66DEFINE_STATIC_KEY_FALSE(switch_to_cond_stibp);
  67/* Control conditional IBPB in switch_mm() */
  68DEFINE_STATIC_KEY_FALSE(switch_mm_cond_ibpb);
  69/* Control unconditional IBPB in switch_mm() */
  70DEFINE_STATIC_KEY_FALSE(switch_mm_always_ibpb);
  71
  72/* Control MDS CPU buffer clear before returning to user space */
  73DEFINE_STATIC_KEY_FALSE(mds_user_clear);
  74EXPORT_SYMBOL_GPL(mds_user_clear);
  75/* Control MDS CPU buffer clear before idling (halt, mwait) */
  76DEFINE_STATIC_KEY_FALSE(mds_idle_clear);
  77EXPORT_SYMBOL_GPL(mds_idle_clear);
  78
  79void __init check_bugs(void)
  80{
  81	identify_boot_cpu();
  82
  83	/*
  84	 * identify_boot_cpu() initialized SMT support information, let the
  85	 * core code know.
  86	 */
  87	cpu_smt_check_topology();
  88
  89	if (!IS_ENABLED(CONFIG_SMP)) {
  90		pr_info("CPU: ");
  91		print_cpu_info(&boot_cpu_data);
  92	}
  93
  94	/*
  95	 * Read the SPEC_CTRL MSR to account for reserved bits which may
  96	 * have unknown values. AMD64_LS_CFG MSR is cached in the early AMD
  97	 * init code as it is not enumerated and depends on the family.
  98	 */
  99	if (boot_cpu_has(X86_FEATURE_MSR_SPEC_CTRL))
 100		rdmsrl(MSR_IA32_SPEC_CTRL, x86_spec_ctrl_base);
 101
 102	/* Allow STIBP in MSR_SPEC_CTRL if supported */
 103	if (boot_cpu_has(X86_FEATURE_STIBP))
 104		x86_spec_ctrl_mask |= SPEC_CTRL_STIBP;
 105
 106	/* Select the proper CPU mitigations before patching alternatives: */
 107	spectre_v1_select_mitigation();
 108	spectre_v2_select_mitigation();
 109	ssb_select_mitigation();
 110	l1tf_select_mitigation();
 111	mds_select_mitigation();
 112	taa_select_mitigation();
 113	srbds_select_mitigation();
 114
 115	/*
 116	 * As MDS and TAA mitigations are inter-related, print MDS
 117	 * mitigation until after TAA mitigation selection is done.
 118	 */
 119	mds_print_mitigation();
 120
 121	arch_smt_update();
 122
 123#ifdef CONFIG_X86_32
 124	/*
 125	 * Check whether we are able to run this kernel safely on SMP.
 126	 *
 127	 * - i386 is no longer supported.
 128	 * - In order to run on anything without a TSC, we need to be
 129	 *   compiled for a i486.
 130	 */
 131	if (boot_cpu_data.x86 < 4)
 132		panic("Kernel requires i486+ for 'invlpg' and other features");
 133
 134	init_utsname()->machine[1] =
 135		'0' + (boot_cpu_data.x86 > 6 ? 6 : boot_cpu_data.x86);
 136	alternative_instructions();
 137
 138	fpu__init_check_bugs();
 139#else /* CONFIG_X86_64 */
 140	alternative_instructions();
 141
 142	/*
 143	 * Make sure the first 2MB area is not mapped by huge pages
 144	 * There are typically fixed size MTRRs in there and overlapping
 145	 * MTRRs into large pages causes slow downs.
 146	 *
 147	 * Right now we don't do that with gbpages because there seems
 148	 * very little benefit for that case.
 149	 */
 150	if (!direct_gbpages)
 151		set_memory_4k((unsigned long)__va(0), 1);
 152#endif
 153}
 154
 155void
 156x86_virt_spec_ctrl(u64 guest_spec_ctrl, u64 guest_virt_spec_ctrl, bool setguest)
 157{
 158	u64 msrval, guestval, hostval = x86_spec_ctrl_base;
 159	struct thread_info *ti = current_thread_info();
 160
 161	/* Is MSR_SPEC_CTRL implemented ? */
 162	if (static_cpu_has(X86_FEATURE_MSR_SPEC_CTRL)) {
 163		/*
 164		 * Restrict guest_spec_ctrl to supported values. Clear the
 165		 * modifiable bits in the host base value and or the
 166		 * modifiable bits from the guest value.
 167		 */
 168		guestval = hostval & ~x86_spec_ctrl_mask;
 169		guestval |= guest_spec_ctrl & x86_spec_ctrl_mask;
 170
 171		/* SSBD controlled in MSR_SPEC_CTRL */
 172		if (static_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) ||
 173		    static_cpu_has(X86_FEATURE_AMD_SSBD))
 174			hostval |= ssbd_tif_to_spec_ctrl(ti->flags);
 175
 176		/* Conditional STIBP enabled? */
 177		if (static_branch_unlikely(&switch_to_cond_stibp))
 178			hostval |= stibp_tif_to_spec_ctrl(ti->flags);
 179
 180		if (hostval != guestval) {
 181			msrval = setguest ? guestval : hostval;
 182			wrmsrl(MSR_IA32_SPEC_CTRL, msrval);
 183		}
 184	}
 185
 186	/*
 187	 * If SSBD is not handled in MSR_SPEC_CTRL on AMD, update
 188	 * MSR_AMD64_L2_CFG or MSR_VIRT_SPEC_CTRL if supported.
 189	 */
 190	if (!static_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
 191	    !static_cpu_has(X86_FEATURE_VIRT_SSBD))
 192		return;
 193
 194	/*
 195	 * If the host has SSBD mitigation enabled, force it in the host's
 196	 * virtual MSR value. If its not permanently enabled, evaluate
 197	 * current's TIF_SSBD thread flag.
 198	 */
 199	if (static_cpu_has(X86_FEATURE_SPEC_STORE_BYPASS_DISABLE))
 200		hostval = SPEC_CTRL_SSBD;
 201	else
 202		hostval = ssbd_tif_to_spec_ctrl(ti->flags);
 203
 204	/* Sanitize the guest value */
 205	guestval = guest_virt_spec_ctrl & SPEC_CTRL_SSBD;
 206
 207	if (hostval != guestval) {
 208		unsigned long tif;
 209
 210		tif = setguest ? ssbd_spec_ctrl_to_tif(guestval) :
 211				 ssbd_spec_ctrl_to_tif(hostval);
 212
 213		speculation_ctrl_update(tif);
 214	}
 215}
 216EXPORT_SYMBOL_GPL(x86_virt_spec_ctrl);
 217
 218static void x86_amd_ssb_disable(void)
 219{
 220	u64 msrval = x86_amd_ls_cfg_base | x86_amd_ls_cfg_ssbd_mask;
 221
 222	if (boot_cpu_has(X86_FEATURE_VIRT_SSBD))
 223		wrmsrl(MSR_AMD64_VIRT_SPEC_CTRL, SPEC_CTRL_SSBD);
 224	else if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD))
 225		wrmsrl(MSR_AMD64_LS_CFG, msrval);
 226}
 227
 228#undef pr_fmt
 229#define pr_fmt(fmt)	"MDS: " fmt
 230
 231/* Default mitigation for MDS-affected CPUs */
 232static enum mds_mitigations mds_mitigation __ro_after_init = MDS_MITIGATION_FULL;
 233static bool mds_nosmt __ro_after_init = false;
 234
 235static const char * const mds_strings[] = {
 236	[MDS_MITIGATION_OFF]	= "Vulnerable",
 237	[MDS_MITIGATION_FULL]	= "Mitigation: Clear CPU buffers",
 238	[MDS_MITIGATION_VMWERV]	= "Vulnerable: Clear CPU buffers attempted, no microcode",
 239};
 240
 241static void __init mds_select_mitigation(void)
 
 
 
 
 
 242{
 243	if (!boot_cpu_has_bug(X86_BUG_MDS) || cpu_mitigations_off()) {
 244		mds_mitigation = MDS_MITIGATION_OFF;
 245		return;
 246	}
 247
 248	if (mds_mitigation == MDS_MITIGATION_FULL) {
 249		if (!boot_cpu_has(X86_FEATURE_MD_CLEAR))
 250			mds_mitigation = MDS_MITIGATION_VMWERV;
 251
 252		static_branch_enable(&mds_user_clear);
 253
 254		if (!boot_cpu_has(X86_BUG_MSBDS_ONLY) &&
 255		    (mds_nosmt || cpu_mitigations_auto_nosmt()))
 256			cpu_smt_disable(false);
 257	}
 258}
 259
 260static void __init mds_print_mitigation(void)
 261{
 262	if (!boot_cpu_has_bug(X86_BUG_MDS) || cpu_mitigations_off())
 263		return;
 264
 265	pr_info("%s\n", mds_strings[mds_mitigation]);
 266}
 267
 268static int __init mds_cmdline(char *str)
 269{
 270	if (!boot_cpu_has_bug(X86_BUG_MDS))
 271		return 0;
 272
 273	if (!str)
 274		return -EINVAL;
 275
 276	if (!strcmp(str, "off"))
 277		mds_mitigation = MDS_MITIGATION_OFF;
 278	else if (!strcmp(str, "full"))
 279		mds_mitigation = MDS_MITIGATION_FULL;
 280	else if (!strcmp(str, "full,nosmt")) {
 281		mds_mitigation = MDS_MITIGATION_FULL;
 282		mds_nosmt = true;
 283	}
 284
 285	return 0;
 286}
 287early_param("mds", mds_cmdline);
 288
 289#undef pr_fmt
 290#define pr_fmt(fmt)	"TAA: " fmt
 291
 292enum taa_mitigations {
 293	TAA_MITIGATION_OFF,
 294	TAA_MITIGATION_UCODE_NEEDED,
 295	TAA_MITIGATION_VERW,
 296	TAA_MITIGATION_TSX_DISABLED,
 297};
 298
 299/* Default mitigation for TAA-affected CPUs */
 300static enum taa_mitigations taa_mitigation __ro_after_init = TAA_MITIGATION_VERW;
 301static bool taa_nosmt __ro_after_init;
 302
 303static const char * const taa_strings[] = {
 304	[TAA_MITIGATION_OFF]		= "Vulnerable",
 305	[TAA_MITIGATION_UCODE_NEEDED]	= "Vulnerable: Clear CPU buffers attempted, no microcode",
 306	[TAA_MITIGATION_VERW]		= "Mitigation: Clear CPU buffers",
 307	[TAA_MITIGATION_TSX_DISABLED]	= "Mitigation: TSX disabled",
 308};
 309
 310static void __init taa_select_mitigation(void)
 311{
 312	u64 ia32_cap;
 313
 314	if (!boot_cpu_has_bug(X86_BUG_TAA)) {
 315		taa_mitigation = TAA_MITIGATION_OFF;
 316		return;
 317	}
 318
 319	/* TSX previously disabled by tsx=off */
 320	if (!boot_cpu_has(X86_FEATURE_RTM)) {
 321		taa_mitigation = TAA_MITIGATION_TSX_DISABLED;
 322		goto out;
 323	}
 324
 325	if (cpu_mitigations_off()) {
 326		taa_mitigation = TAA_MITIGATION_OFF;
 327		return;
 328	}
 329
 330	/*
 331	 * TAA mitigation via VERW is turned off if both
 332	 * tsx_async_abort=off and mds=off are specified.
 333	 */
 334	if (taa_mitigation == TAA_MITIGATION_OFF &&
 335	    mds_mitigation == MDS_MITIGATION_OFF)
 336		goto out;
 337
 338	if (boot_cpu_has(X86_FEATURE_MD_CLEAR))
 339		taa_mitigation = TAA_MITIGATION_VERW;
 340	else
 341		taa_mitigation = TAA_MITIGATION_UCODE_NEEDED;
 342
 343	/*
 344	 * VERW doesn't clear the CPU buffers when MD_CLEAR=1 and MDS_NO=1.
 345	 * A microcode update fixes this behavior to clear CPU buffers. It also
 346	 * adds support for MSR_IA32_TSX_CTRL which is enumerated by the
 347	 * ARCH_CAP_TSX_CTRL_MSR bit.
 348	 *
 349	 * On MDS_NO=1 CPUs if ARCH_CAP_TSX_CTRL_MSR is not set, microcode
 350	 * update is required.
 351	 */
 352	ia32_cap = x86_read_arch_cap_msr();
 353	if ( (ia32_cap & ARCH_CAP_MDS_NO) &&
 354	    !(ia32_cap & ARCH_CAP_TSX_CTRL_MSR))
 355		taa_mitigation = TAA_MITIGATION_UCODE_NEEDED;
 356
 357	/*
 358	 * TSX is enabled, select alternate mitigation for TAA which is
 359	 * the same as MDS. Enable MDS static branch to clear CPU buffers.
 360	 *
 361	 * For guests that can't determine whether the correct microcode is
 362	 * present on host, enable the mitigation for UCODE_NEEDED as well.
 363	 */
 364	static_branch_enable(&mds_user_clear);
 
 
 
 
 
 
 
 
 
 
 365
 366	if (taa_nosmt || cpu_mitigations_auto_nosmt())
 367		cpu_smt_disable(false);
 368
 369	/*
 370	 * Update MDS mitigation, if necessary, as the mds_user_clear is
 371	 * now enabled for TAA mitigation.
 372	 */
 373	if (mds_mitigation == MDS_MITIGATION_OFF &&
 374	    boot_cpu_has_bug(X86_BUG_MDS)) {
 375		mds_mitigation = MDS_MITIGATION_FULL;
 376		mds_select_mitigation();
 377	}
 378out:
 379	pr_info("%s\n", taa_strings[taa_mitigation]);
 380}
 381
 382static int __init tsx_async_abort_parse_cmdline(char *str)
 383{
 384	if (!boot_cpu_has_bug(X86_BUG_TAA))
 385		return 0;
 386
 387	if (!str)
 388		return -EINVAL;
 389
 390	if (!strcmp(str, "off")) {
 391		taa_mitigation = TAA_MITIGATION_OFF;
 392	} else if (!strcmp(str, "full")) {
 393		taa_mitigation = TAA_MITIGATION_VERW;
 394	} else if (!strcmp(str, "full,nosmt")) {
 395		taa_mitigation = TAA_MITIGATION_VERW;
 396		taa_nosmt = true;
 397	}
 398
 399	return 0;
 400}
 401early_param("tsx_async_abort", tsx_async_abort_parse_cmdline);
 402
 403#undef pr_fmt
 404#define pr_fmt(fmt)	"SRBDS: " fmt
 405
 406enum srbds_mitigations {
 407	SRBDS_MITIGATION_OFF,
 408	SRBDS_MITIGATION_UCODE_NEEDED,
 409	SRBDS_MITIGATION_FULL,
 410	SRBDS_MITIGATION_TSX_OFF,
 411	SRBDS_MITIGATION_HYPERVISOR,
 412};
 413
 414static enum srbds_mitigations srbds_mitigation __ro_after_init = SRBDS_MITIGATION_FULL;
 415
 416static const char * const srbds_strings[] = {
 417	[SRBDS_MITIGATION_OFF]		= "Vulnerable",
 418	[SRBDS_MITIGATION_UCODE_NEEDED]	= "Vulnerable: No microcode",
 419	[SRBDS_MITIGATION_FULL]		= "Mitigation: Microcode",
 420	[SRBDS_MITIGATION_TSX_OFF]	= "Mitigation: TSX disabled",
 421	[SRBDS_MITIGATION_HYPERVISOR]	= "Unknown: Dependent on hypervisor status",
 422};
 423
 424static bool srbds_off;
 425
 426void update_srbds_msr(void)
 427{
 428	u64 mcu_ctrl;
 429
 430	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
 431		return;
 432
 433	if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
 434		return;
 435
 436	if (srbds_mitigation == SRBDS_MITIGATION_UCODE_NEEDED)
 
 
 437		return;
 438
 439	rdmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
 440
 441	switch (srbds_mitigation) {
 442	case SRBDS_MITIGATION_OFF:
 443	case SRBDS_MITIGATION_TSX_OFF:
 444		mcu_ctrl |= RNGDS_MITG_DIS;
 445		break;
 446	case SRBDS_MITIGATION_FULL:
 447		mcu_ctrl &= ~RNGDS_MITG_DIS;
 448		break;
 449	default:
 450		break;
 451	}
 452
 453	wrmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
 454}
 455
 456static void __init srbds_select_mitigation(void)
 457{
 458	u64 ia32_cap;
 459
 460	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
 461		return;
 462
 463	/*
 464	 * Check to see if this is one of the MDS_NO systems supporting
 465	 * TSX that are only exposed to SRBDS when TSX is enabled.
 466	 */
 467	ia32_cap = x86_read_arch_cap_msr();
 468	if ((ia32_cap & ARCH_CAP_MDS_NO) && !boot_cpu_has(X86_FEATURE_RTM))
 469		srbds_mitigation = SRBDS_MITIGATION_TSX_OFF;
 470	else if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
 471		srbds_mitigation = SRBDS_MITIGATION_HYPERVISOR;
 472	else if (!boot_cpu_has(X86_FEATURE_SRBDS_CTRL))
 473		srbds_mitigation = SRBDS_MITIGATION_UCODE_NEEDED;
 474	else if (cpu_mitigations_off() || srbds_off)
 475		srbds_mitigation = SRBDS_MITIGATION_OFF;
 476
 477	update_srbds_msr();
 478	pr_info("%s\n", srbds_strings[srbds_mitigation]);
 479}
 480
 481static int __init srbds_parse_cmdline(char *str)
 482{
 483	if (!str)
 484		return -EINVAL;
 485
 486	if (!boot_cpu_has_bug(X86_BUG_SRBDS))
 487		return 0;
 488
 489	srbds_off = !strcmp(str, "off");
 490	return 0;
 491}
 492early_param("srbds", srbds_parse_cmdline);
 493
 494#undef pr_fmt
 495#define pr_fmt(fmt)     "Spectre V1 : " fmt
 496
 497enum spectre_v1_mitigation {
 498	SPECTRE_V1_MITIGATION_NONE,
 499	SPECTRE_V1_MITIGATION_AUTO,
 500};
 501
 502static enum spectre_v1_mitigation spectre_v1_mitigation __ro_after_init =
 503	SPECTRE_V1_MITIGATION_AUTO;
 504
 505static const char * const spectre_v1_strings[] = {
 506	[SPECTRE_V1_MITIGATION_NONE] = "Vulnerable: __user pointer sanitization and usercopy barriers only; no swapgs barriers",
 507	[SPECTRE_V1_MITIGATION_AUTO] = "Mitigation: usercopy/swapgs barriers and __user pointer sanitization",
 508};
 509
 510/*
 511 * Does SMAP provide full mitigation against speculative kernel access to
 512 * userspace?
 513 */
 514static bool smap_works_speculatively(void)
 515{
 516	if (!boot_cpu_has(X86_FEATURE_SMAP))
 517		return false;
 518
 519	/*
 520	 * On CPUs which are vulnerable to Meltdown, SMAP does not
 521	 * prevent speculative access to user data in the L1 cache.
 522	 * Consider SMAP to be non-functional as a mitigation on these
 523	 * CPUs.
 524	 */
 525	if (boot_cpu_has(X86_BUG_CPU_MELTDOWN))
 526		return false;
 527
 528	return true;
 529}
 530
 531static void __init spectre_v1_select_mitigation(void)
 532{
 533	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V1) || cpu_mitigations_off()) {
 534		spectre_v1_mitigation = SPECTRE_V1_MITIGATION_NONE;
 535		return;
 536	}
 537
 538	if (spectre_v1_mitigation == SPECTRE_V1_MITIGATION_AUTO) {
 539		/*
 540		 * With Spectre v1, a user can speculatively control either
 541		 * path of a conditional swapgs with a user-controlled GS
 542		 * value.  The mitigation is to add lfences to both code paths.
 543		 *
 544		 * If FSGSBASE is enabled, the user can put a kernel address in
 545		 * GS, in which case SMAP provides no protection.
 546		 *
 547		 * If FSGSBASE is disabled, the user can only put a user space
 548		 * address in GS.  That makes an attack harder, but still
 549		 * possible if there's no SMAP protection.
 550		 */
 551		if (boot_cpu_has(X86_FEATURE_FSGSBASE) ||
 552		    !smap_works_speculatively()) {
 553			/*
 554			 * Mitigation can be provided from SWAPGS itself or
 555			 * PTI as the CR3 write in the Meltdown mitigation
 556			 * is serializing.
 557			 *
 558			 * If neither is there, mitigate with an LFENCE to
 559			 * stop speculation through swapgs.
 560			 */
 561			if (boot_cpu_has_bug(X86_BUG_SWAPGS) &&
 562			    !boot_cpu_has(X86_FEATURE_PTI))
 563				setup_force_cpu_cap(X86_FEATURE_FENCE_SWAPGS_USER);
 564
 565			/*
 566			 * Enable lfences in the kernel entry (non-swapgs)
 567			 * paths, to prevent user entry from speculatively
 568			 * skipping swapgs.
 569			 */
 570			setup_force_cpu_cap(X86_FEATURE_FENCE_SWAPGS_KERNEL);
 571		}
 572	}
 573
 574	pr_info("%s\n", spectre_v1_strings[spectre_v1_mitigation]);
 575}
 576
 577static int __init nospectre_v1_cmdline(char *str)
 578{
 579	spectre_v1_mitigation = SPECTRE_V1_MITIGATION_NONE;
 580	return 0;
 581}
 582early_param("nospectre_v1", nospectre_v1_cmdline);
 583
 584#undef pr_fmt
 585#define pr_fmt(fmt)     "Spectre V2 : " fmt
 586
 587static enum spectre_v2_mitigation spectre_v2_enabled __ro_after_init =
 588	SPECTRE_V2_NONE;
 589
 590static enum spectre_v2_user_mitigation spectre_v2_user_stibp __ro_after_init =
 591	SPECTRE_V2_USER_NONE;
 592static enum spectre_v2_user_mitigation spectre_v2_user_ibpb __ro_after_init =
 593	SPECTRE_V2_USER_NONE;
 594
 595#ifdef CONFIG_RETPOLINE
 596static bool spectre_v2_bad_module;
 597
 598bool retpoline_module_ok(bool has_retpoline)
 599{
 600	if (spectre_v2_enabled == SPECTRE_V2_NONE || has_retpoline)
 601		return true;
 602
 603	pr_err("System may be vulnerable to spectre v2\n");
 604	spectre_v2_bad_module = true;
 605	return false;
 606}
 607
 608static inline const char *spectre_v2_module_string(void)
 609{
 610	return spectre_v2_bad_module ? " - vulnerable module loaded" : "";
 611}
 612#else
 613static inline const char *spectre_v2_module_string(void) { return ""; }
 614#endif
 615
 616static inline bool match_option(const char *arg, int arglen, const char *opt)
 617{
 618	int len = strlen(opt);
 619
 620	return len == arglen && !strncmp(arg, opt, len);
 621}
 622
 623/* The kernel command line selection for spectre v2 */
 624enum spectre_v2_mitigation_cmd {
 625	SPECTRE_V2_CMD_NONE,
 626	SPECTRE_V2_CMD_AUTO,
 627	SPECTRE_V2_CMD_FORCE,
 628	SPECTRE_V2_CMD_RETPOLINE,
 629	SPECTRE_V2_CMD_RETPOLINE_GENERIC,
 630	SPECTRE_V2_CMD_RETPOLINE_AMD,
 631};
 632
 633enum spectre_v2_user_cmd {
 634	SPECTRE_V2_USER_CMD_NONE,
 635	SPECTRE_V2_USER_CMD_AUTO,
 636	SPECTRE_V2_USER_CMD_FORCE,
 637	SPECTRE_V2_USER_CMD_PRCTL,
 638	SPECTRE_V2_USER_CMD_PRCTL_IBPB,
 639	SPECTRE_V2_USER_CMD_SECCOMP,
 640	SPECTRE_V2_USER_CMD_SECCOMP_IBPB,
 641};
 642
 643static const char * const spectre_v2_user_strings[] = {
 644	[SPECTRE_V2_USER_NONE]			= "User space: Vulnerable",
 645	[SPECTRE_V2_USER_STRICT]		= "User space: Mitigation: STIBP protection",
 646	[SPECTRE_V2_USER_STRICT_PREFERRED]	= "User space: Mitigation: STIBP always-on protection",
 647	[SPECTRE_V2_USER_PRCTL]			= "User space: Mitigation: STIBP via prctl",
 648	[SPECTRE_V2_USER_SECCOMP]		= "User space: Mitigation: STIBP via seccomp and prctl",
 649};
 650
 651static const struct {
 652	const char			*option;
 653	enum spectre_v2_user_cmd	cmd;
 654	bool				secure;
 655} v2_user_options[] __initconst = {
 656	{ "auto",		SPECTRE_V2_USER_CMD_AUTO,		false },
 657	{ "off",		SPECTRE_V2_USER_CMD_NONE,		false },
 658	{ "on",			SPECTRE_V2_USER_CMD_FORCE,		true  },
 659	{ "prctl",		SPECTRE_V2_USER_CMD_PRCTL,		false },
 660	{ "prctl,ibpb",		SPECTRE_V2_USER_CMD_PRCTL_IBPB,		false },
 661	{ "seccomp",		SPECTRE_V2_USER_CMD_SECCOMP,		false },
 662	{ "seccomp,ibpb",	SPECTRE_V2_USER_CMD_SECCOMP_IBPB,	false },
 663};
 664
 665static void __init spec_v2_user_print_cond(const char *reason, bool secure)
 666{
 667	if (boot_cpu_has_bug(X86_BUG_SPECTRE_V2) != secure)
 668		pr_info("spectre_v2_user=%s forced on command line.\n", reason);
 669}
 670
 671static enum spectre_v2_user_cmd __init
 672spectre_v2_parse_user_cmdline(enum spectre_v2_mitigation_cmd v2_cmd)
 673{
 674	char arg[20];
 675	int ret, i;
 676
 677	switch (v2_cmd) {
 678	case SPECTRE_V2_CMD_NONE:
 679		return SPECTRE_V2_USER_CMD_NONE;
 680	case SPECTRE_V2_CMD_FORCE:
 681		return SPECTRE_V2_USER_CMD_FORCE;
 682	default:
 683		break;
 684	}
 685
 686	ret = cmdline_find_option(boot_command_line, "spectre_v2_user",
 687				  arg, sizeof(arg));
 688	if (ret < 0)
 689		return SPECTRE_V2_USER_CMD_AUTO;
 690
 691	for (i = 0; i < ARRAY_SIZE(v2_user_options); i++) {
 692		if (match_option(arg, ret, v2_user_options[i].option)) {
 693			spec_v2_user_print_cond(v2_user_options[i].option,
 694						v2_user_options[i].secure);
 695			return v2_user_options[i].cmd;
 696		}
 697	}
 698
 699	pr_err("Unknown user space protection option (%s). Switching to AUTO select\n", arg);
 700	return SPECTRE_V2_USER_CMD_AUTO;
 701}
 702
 703static void __init
 704spectre_v2_user_select_mitigation(enum spectre_v2_mitigation_cmd v2_cmd)
 705{
 706	enum spectre_v2_user_mitigation mode = SPECTRE_V2_USER_NONE;
 707	bool smt_possible = IS_ENABLED(CONFIG_SMP);
 708	enum spectre_v2_user_cmd cmd;
 709
 710	if (!boot_cpu_has(X86_FEATURE_IBPB) && !boot_cpu_has(X86_FEATURE_STIBP))
 711		return;
 712
 713	if (cpu_smt_control == CPU_SMT_FORCE_DISABLED ||
 714	    cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
 715		smt_possible = false;
 716
 717	cmd = spectre_v2_parse_user_cmdline(v2_cmd);
 718	switch (cmd) {
 719	case SPECTRE_V2_USER_CMD_NONE:
 720		goto set_mode;
 721	case SPECTRE_V2_USER_CMD_FORCE:
 722		mode = SPECTRE_V2_USER_STRICT;
 723		break;
 724	case SPECTRE_V2_USER_CMD_PRCTL:
 725	case SPECTRE_V2_USER_CMD_PRCTL_IBPB:
 726		mode = SPECTRE_V2_USER_PRCTL;
 727		break;
 728	case SPECTRE_V2_USER_CMD_AUTO:
 729	case SPECTRE_V2_USER_CMD_SECCOMP:
 730	case SPECTRE_V2_USER_CMD_SECCOMP_IBPB:
 731		if (IS_ENABLED(CONFIG_SECCOMP))
 732			mode = SPECTRE_V2_USER_SECCOMP;
 733		else
 734			mode = SPECTRE_V2_USER_PRCTL;
 735		break;
 736	}
 737
 738	/* Initialize Indirect Branch Prediction Barrier */
 739	if (boot_cpu_has(X86_FEATURE_IBPB)) {
 740		setup_force_cpu_cap(X86_FEATURE_USE_IBPB);
 741
 742		spectre_v2_user_ibpb = mode;
 743		switch (cmd) {
 744		case SPECTRE_V2_USER_CMD_FORCE:
 745		case SPECTRE_V2_USER_CMD_PRCTL_IBPB:
 746		case SPECTRE_V2_USER_CMD_SECCOMP_IBPB:
 747			static_branch_enable(&switch_mm_always_ibpb);
 748			spectre_v2_user_ibpb = SPECTRE_V2_USER_STRICT;
 749			break;
 750		case SPECTRE_V2_USER_CMD_PRCTL:
 751		case SPECTRE_V2_USER_CMD_AUTO:
 752		case SPECTRE_V2_USER_CMD_SECCOMP:
 753			static_branch_enable(&switch_mm_cond_ibpb);
 754			break;
 755		default:
 756			break;
 757		}
 758
 759		pr_info("mitigation: Enabling %s Indirect Branch Prediction Barrier\n",
 760			static_key_enabled(&switch_mm_always_ibpb) ?
 761			"always-on" : "conditional");
 762	}
 763
 
 
 
 
 
 
 764	/*
 765	 * If no STIBP, enhanced IBRS is enabled or SMT impossible, STIBP is not
 766	 * required.
 767	 */
 768	if (!boot_cpu_has(X86_FEATURE_STIBP) ||
 769	    !smt_possible ||
 770	    spectre_v2_enabled == SPECTRE_V2_IBRS_ENHANCED)
 771		return;
 772
 773	/*
 774	 * At this point, an STIBP mode other than "off" has been set.
 775	 * If STIBP support is not being forced, check if STIBP always-on
 776	 * is preferred.
 777	 */
 778	if (mode != SPECTRE_V2_USER_STRICT &&
 779	    boot_cpu_has(X86_FEATURE_AMD_STIBP_ALWAYS_ON))
 780		mode = SPECTRE_V2_USER_STRICT_PREFERRED;
 781
 782	spectre_v2_user_stibp = mode;
 783
 784set_mode:
 785	pr_info("%s\n", spectre_v2_user_strings[mode]);
 786}
 787
 788static const char * const spectre_v2_strings[] = {
 789	[SPECTRE_V2_NONE]			= "Vulnerable",
 790	[SPECTRE_V2_RETPOLINE_GENERIC]		= "Mitigation: Full generic retpoline",
 791	[SPECTRE_V2_RETPOLINE_AMD]		= "Mitigation: Full AMD retpoline",
 792	[SPECTRE_V2_IBRS_ENHANCED]		= "Mitigation: Enhanced IBRS",
 793};
 794
 795static const struct {
 796	const char *option;
 797	enum spectre_v2_mitigation_cmd cmd;
 798	bool secure;
 799} mitigation_options[] __initconst = {
 800	{ "off",		SPECTRE_V2_CMD_NONE,		  false },
 801	{ "on",			SPECTRE_V2_CMD_FORCE,		  true  },
 802	{ "retpoline",		SPECTRE_V2_CMD_RETPOLINE,	  false },
 803	{ "retpoline,amd",	SPECTRE_V2_CMD_RETPOLINE_AMD,	  false },
 804	{ "retpoline,generic",	SPECTRE_V2_CMD_RETPOLINE_GENERIC, false },
 805	{ "auto",		SPECTRE_V2_CMD_AUTO,		  false },
 806};
 807
 808static void __init spec_v2_print_cond(const char *reason, bool secure)
 809{
 810	if (boot_cpu_has_bug(X86_BUG_SPECTRE_V2) != secure)
 811		pr_info("%s selected on command line.\n", reason);
 812}
 813
 814static enum spectre_v2_mitigation_cmd __init spectre_v2_parse_cmdline(void)
 815{
 816	enum spectre_v2_mitigation_cmd cmd = SPECTRE_V2_CMD_AUTO;
 817	char arg[20];
 818	int ret, i;
 819
 820	if (cmdline_find_option_bool(boot_command_line, "nospectre_v2") ||
 821	    cpu_mitigations_off())
 822		return SPECTRE_V2_CMD_NONE;
 823
 824	ret = cmdline_find_option(boot_command_line, "spectre_v2", arg, sizeof(arg));
 825	if (ret < 0)
 826		return SPECTRE_V2_CMD_AUTO;
 827
 828	for (i = 0; i < ARRAY_SIZE(mitigation_options); i++) {
 829		if (!match_option(arg, ret, mitigation_options[i].option))
 830			continue;
 831		cmd = mitigation_options[i].cmd;
 832		break;
 833	}
 834
 835	if (i >= ARRAY_SIZE(mitigation_options)) {
 836		pr_err("unknown option (%s). Switching to AUTO select\n", arg);
 837		return SPECTRE_V2_CMD_AUTO;
 838	}
 839
 840	if ((cmd == SPECTRE_V2_CMD_RETPOLINE ||
 841	     cmd == SPECTRE_V2_CMD_RETPOLINE_AMD ||
 842	     cmd == SPECTRE_V2_CMD_RETPOLINE_GENERIC) &&
 843	    !IS_ENABLED(CONFIG_RETPOLINE)) {
 844		pr_err("%s selected but not compiled in. Switching to AUTO select\n", mitigation_options[i].option);
 845		return SPECTRE_V2_CMD_AUTO;
 846	}
 847
 848	if (cmd == SPECTRE_V2_CMD_RETPOLINE_AMD &&
 849	    boot_cpu_data.x86_vendor != X86_VENDOR_HYGON &&
 850	    boot_cpu_data.x86_vendor != X86_VENDOR_AMD) {
 851		pr_err("retpoline,amd selected but CPU is not AMD. Switching to AUTO select\n");
 852		return SPECTRE_V2_CMD_AUTO;
 853	}
 854
 855	spec_v2_print_cond(mitigation_options[i].option,
 856			   mitigation_options[i].secure);
 857	return cmd;
 858}
 859
 860static void __init spectre_v2_select_mitigation(void)
 861{
 862	enum spectre_v2_mitigation_cmd cmd = spectre_v2_parse_cmdline();
 863	enum spectre_v2_mitigation mode = SPECTRE_V2_NONE;
 864
 865	/*
 866	 * If the CPU is not affected and the command line mode is NONE or AUTO
 867	 * then nothing to do.
 868	 */
 869	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V2) &&
 870	    (cmd == SPECTRE_V2_CMD_NONE || cmd == SPECTRE_V2_CMD_AUTO))
 871		return;
 872
 873	switch (cmd) {
 874	case SPECTRE_V2_CMD_NONE:
 875		return;
 876
 877	case SPECTRE_V2_CMD_FORCE:
 878	case SPECTRE_V2_CMD_AUTO:
 879		if (boot_cpu_has(X86_FEATURE_IBRS_ENHANCED)) {
 880			mode = SPECTRE_V2_IBRS_ENHANCED;
 881			/* Force it so VMEXIT will restore correctly */
 882			x86_spec_ctrl_base |= SPEC_CTRL_IBRS;
 883			wrmsrl(MSR_IA32_SPEC_CTRL, x86_spec_ctrl_base);
 884			goto specv2_set_mode;
 885		}
 886		if (IS_ENABLED(CONFIG_RETPOLINE))
 887			goto retpoline_auto;
 888		break;
 889	case SPECTRE_V2_CMD_RETPOLINE_AMD:
 890		if (IS_ENABLED(CONFIG_RETPOLINE))
 891			goto retpoline_amd;
 892		break;
 893	case SPECTRE_V2_CMD_RETPOLINE_GENERIC:
 894		if (IS_ENABLED(CONFIG_RETPOLINE))
 895			goto retpoline_generic;
 896		break;
 897	case SPECTRE_V2_CMD_RETPOLINE:
 898		if (IS_ENABLED(CONFIG_RETPOLINE))
 899			goto retpoline_auto;
 900		break;
 901	}
 902	pr_err("Spectre mitigation: kernel not compiled with retpoline; no mitigation available!");
 903	return;
 904
 905retpoline_auto:
 906	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
 907	    boot_cpu_data.x86_vendor == X86_VENDOR_HYGON) {
 908	retpoline_amd:
 909		if (!boot_cpu_has(X86_FEATURE_LFENCE_RDTSC)) {
 910			pr_err("Spectre mitigation: LFENCE not serializing, switching to generic retpoline\n");
 911			goto retpoline_generic;
 912		}
 913		mode = SPECTRE_V2_RETPOLINE_AMD;
 914		setup_force_cpu_cap(X86_FEATURE_RETPOLINE_AMD);
 915		setup_force_cpu_cap(X86_FEATURE_RETPOLINE);
 916	} else {
 917	retpoline_generic:
 918		mode = SPECTRE_V2_RETPOLINE_GENERIC;
 919		setup_force_cpu_cap(X86_FEATURE_RETPOLINE);
 920	}
 921
 922specv2_set_mode:
 923	spectre_v2_enabled = mode;
 924	pr_info("%s\n", spectre_v2_strings[mode]);
 925
 926	/*
 927	 * If spectre v2 protection has been enabled, unconditionally fill
 928	 * RSB during a context switch; this protects against two independent
 929	 * issues:
 930	 *
 931	 *	- RSB underflow (and switch to BTB) on Skylake+
 932	 *	- SpectreRSB variant of spectre v2 on X86_BUG_SPECTRE_V2 CPUs
 933	 */
 934	setup_force_cpu_cap(X86_FEATURE_RSB_CTXSW);
 935	pr_info("Spectre v2 / SpectreRSB mitigation: Filling RSB on context switch\n");
 936
 937	/*
 938	 * Retpoline means the kernel is safe because it has no indirect
 939	 * branches. Enhanced IBRS protects firmware too, so, enable restricted
 940	 * speculation around firmware calls only when Enhanced IBRS isn't
 941	 * supported.
 942	 *
 943	 * Use "mode" to check Enhanced IBRS instead of boot_cpu_has(), because
 944	 * the user might select retpoline on the kernel command line and if
 945	 * the CPU supports Enhanced IBRS, kernel might un-intentionally not
 946	 * enable IBRS around firmware calls.
 947	 */
 948	if (boot_cpu_has(X86_FEATURE_IBRS) && mode != SPECTRE_V2_IBRS_ENHANCED) {
 949		setup_force_cpu_cap(X86_FEATURE_USE_IBRS_FW);
 950		pr_info("Enabling Restricted Speculation for firmware calls\n");
 951	}
 952
 953	/* Set up IBPB and STIBP depending on the general spectre V2 command */
 954	spectre_v2_user_select_mitigation(cmd);
 955}
 956
 957static void update_stibp_msr(void * __unused)
 958{
 959	wrmsrl(MSR_IA32_SPEC_CTRL, x86_spec_ctrl_base);
 960}
 961
 962/* Update x86_spec_ctrl_base in case SMT state changed. */
 963static void update_stibp_strict(void)
 964{
 965	u64 mask = x86_spec_ctrl_base & ~SPEC_CTRL_STIBP;
 966
 967	if (sched_smt_active())
 968		mask |= SPEC_CTRL_STIBP;
 969
 970	if (mask == x86_spec_ctrl_base)
 971		return;
 972
 973	pr_info("Update user space SMT mitigation: STIBP %s\n",
 974		mask & SPEC_CTRL_STIBP ? "always-on" : "off");
 975	x86_spec_ctrl_base = mask;
 976	on_each_cpu(update_stibp_msr, NULL, 1);
 977}
 978
 979/* Update the static key controlling the evaluation of TIF_SPEC_IB */
 980static void update_indir_branch_cond(void)
 981{
 982	if (sched_smt_active())
 983		static_branch_enable(&switch_to_cond_stibp);
 984	else
 985		static_branch_disable(&switch_to_cond_stibp);
 986}
 987
 988#undef pr_fmt
 989#define pr_fmt(fmt) fmt
 990
 991/* Update the static key controlling the MDS CPU buffer clear in idle */
 992static void update_mds_branch_idle(void)
 993{
 994	/*
 995	 * Enable the idle clearing if SMT is active on CPUs which are
 996	 * affected only by MSBDS and not any other MDS variant.
 997	 *
 998	 * The other variants cannot be mitigated when SMT is enabled, so
 999	 * clearing the buffers on idle just to prevent the Store Buffer
1000	 * repartitioning leak would be a window dressing exercise.
1001	 */
1002	if (!boot_cpu_has_bug(X86_BUG_MSBDS_ONLY))
1003		return;
1004
1005	if (sched_smt_active())
1006		static_branch_enable(&mds_idle_clear);
1007	else
1008		static_branch_disable(&mds_idle_clear);
1009}
1010
1011#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"
1012#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"
1013
1014void cpu_bugs_smt_update(void)
1015{
1016	mutex_lock(&spec_ctrl_mutex);
1017
1018	switch (spectre_v2_user_stibp) {
1019	case SPECTRE_V2_USER_NONE:
1020		break;
1021	case SPECTRE_V2_USER_STRICT:
1022	case SPECTRE_V2_USER_STRICT_PREFERRED:
1023		update_stibp_strict();
1024		break;
1025	case SPECTRE_V2_USER_PRCTL:
1026	case SPECTRE_V2_USER_SECCOMP:
1027		update_indir_branch_cond();
1028		break;
1029	}
1030
1031	switch (mds_mitigation) {
1032	case MDS_MITIGATION_FULL:
1033	case MDS_MITIGATION_VMWERV:
1034		if (sched_smt_active() && !boot_cpu_has(X86_BUG_MSBDS_ONLY))
1035			pr_warn_once(MDS_MSG_SMT);
1036		update_mds_branch_idle();
1037		break;
1038	case MDS_MITIGATION_OFF:
1039		break;
1040	}
1041
1042	switch (taa_mitigation) {
1043	case TAA_MITIGATION_VERW:
1044	case TAA_MITIGATION_UCODE_NEEDED:
1045		if (sched_smt_active())
1046			pr_warn_once(TAA_MSG_SMT);
1047		break;
1048	case TAA_MITIGATION_TSX_DISABLED:
1049	case TAA_MITIGATION_OFF:
1050		break;
1051	}
1052
1053	mutex_unlock(&spec_ctrl_mutex);
1054}
1055
1056#undef pr_fmt
1057#define pr_fmt(fmt)	"Speculative Store Bypass: " fmt
1058
1059static enum ssb_mitigation ssb_mode __ro_after_init = SPEC_STORE_BYPASS_NONE;
1060
1061/* The kernel command line selection */
1062enum ssb_mitigation_cmd {
1063	SPEC_STORE_BYPASS_CMD_NONE,
1064	SPEC_STORE_BYPASS_CMD_AUTO,
1065	SPEC_STORE_BYPASS_CMD_ON,
1066	SPEC_STORE_BYPASS_CMD_PRCTL,
1067	SPEC_STORE_BYPASS_CMD_SECCOMP,
1068};
1069
1070static const char * const ssb_strings[] = {
1071	[SPEC_STORE_BYPASS_NONE]	= "Vulnerable",
1072	[SPEC_STORE_BYPASS_DISABLE]	= "Mitigation: Speculative Store Bypass disabled",
1073	[SPEC_STORE_BYPASS_PRCTL]	= "Mitigation: Speculative Store Bypass disabled via prctl",
1074	[SPEC_STORE_BYPASS_SECCOMP]	= "Mitigation: Speculative Store Bypass disabled via prctl and seccomp",
1075};
1076
1077static const struct {
1078	const char *option;
1079	enum ssb_mitigation_cmd cmd;
1080} ssb_mitigation_options[]  __initconst = {
1081	{ "auto",	SPEC_STORE_BYPASS_CMD_AUTO },    /* Platform decides */
1082	{ "on",		SPEC_STORE_BYPASS_CMD_ON },      /* Disable Speculative Store Bypass */
1083	{ "off",	SPEC_STORE_BYPASS_CMD_NONE },    /* Don't touch Speculative Store Bypass */
1084	{ "prctl",	SPEC_STORE_BYPASS_CMD_PRCTL },   /* Disable Speculative Store Bypass via prctl */
1085	{ "seccomp",	SPEC_STORE_BYPASS_CMD_SECCOMP }, /* Disable Speculative Store Bypass via prctl and seccomp */
1086};
1087
1088static enum ssb_mitigation_cmd __init ssb_parse_cmdline(void)
1089{
1090	enum ssb_mitigation_cmd cmd = SPEC_STORE_BYPASS_CMD_AUTO;
1091	char arg[20];
1092	int ret, i;
1093
1094	if (cmdline_find_option_bool(boot_command_line, "nospec_store_bypass_disable") ||
1095	    cpu_mitigations_off()) {
1096		return SPEC_STORE_BYPASS_CMD_NONE;
1097	} else {
1098		ret = cmdline_find_option(boot_command_line, "spec_store_bypass_disable",
1099					  arg, sizeof(arg));
1100		if (ret < 0)
1101			return SPEC_STORE_BYPASS_CMD_AUTO;
1102
1103		for (i = 0; i < ARRAY_SIZE(ssb_mitigation_options); i++) {
1104			if (!match_option(arg, ret, ssb_mitigation_options[i].option))
1105				continue;
1106
1107			cmd = ssb_mitigation_options[i].cmd;
1108			break;
1109		}
1110
1111		if (i >= ARRAY_SIZE(ssb_mitigation_options)) {
1112			pr_err("unknown option (%s). Switching to AUTO select\n", arg);
1113			return SPEC_STORE_BYPASS_CMD_AUTO;
1114		}
1115	}
1116
1117	return cmd;
1118}
1119
1120static enum ssb_mitigation __init __ssb_select_mitigation(void)
1121{
1122	enum ssb_mitigation mode = SPEC_STORE_BYPASS_NONE;
1123	enum ssb_mitigation_cmd cmd;
1124
1125	if (!boot_cpu_has(X86_FEATURE_SSBD))
1126		return mode;
1127
1128	cmd = ssb_parse_cmdline();
1129	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS) &&
1130	    (cmd == SPEC_STORE_BYPASS_CMD_NONE ||
1131	     cmd == SPEC_STORE_BYPASS_CMD_AUTO))
1132		return mode;
1133
1134	switch (cmd) {
1135	case SPEC_STORE_BYPASS_CMD_AUTO:
1136	case SPEC_STORE_BYPASS_CMD_SECCOMP:
1137		/*
1138		 * Choose prctl+seccomp as the default mode if seccomp is
1139		 * enabled.
1140		 */
1141		if (IS_ENABLED(CONFIG_SECCOMP))
1142			mode = SPEC_STORE_BYPASS_SECCOMP;
1143		else
1144			mode = SPEC_STORE_BYPASS_PRCTL;
1145		break;
1146	case SPEC_STORE_BYPASS_CMD_ON:
1147		mode = SPEC_STORE_BYPASS_DISABLE;
1148		break;
1149	case SPEC_STORE_BYPASS_CMD_PRCTL:
1150		mode = SPEC_STORE_BYPASS_PRCTL;
1151		break;
1152	case SPEC_STORE_BYPASS_CMD_NONE:
1153		break;
1154	}
1155
1156	/*
1157	 * If SSBD is controlled by the SPEC_CTRL MSR, then set the proper
1158	 * bit in the mask to allow guests to use the mitigation even in the
1159	 * case where the host does not enable it.
1160	 */
1161	if (static_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) ||
1162	    static_cpu_has(X86_FEATURE_AMD_SSBD)) {
1163		x86_spec_ctrl_mask |= SPEC_CTRL_SSBD;
1164	}
1165
1166	/*
1167	 * We have three CPU feature flags that are in play here:
1168	 *  - X86_BUG_SPEC_STORE_BYPASS - CPU is susceptible.
1169	 *  - X86_FEATURE_SSBD - CPU is able to turn off speculative store bypass
1170	 *  - X86_FEATURE_SPEC_STORE_BYPASS_DISABLE - engage the mitigation
1171	 */
1172	if (mode == SPEC_STORE_BYPASS_DISABLE) {
1173		setup_force_cpu_cap(X86_FEATURE_SPEC_STORE_BYPASS_DISABLE);
1174		/*
1175		 * Intel uses the SPEC CTRL MSR Bit(2) for this, while AMD may
1176		 * use a completely different MSR and bit dependent on family.
1177		 */
1178		if (!static_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) &&
1179		    !static_cpu_has(X86_FEATURE_AMD_SSBD)) {
1180			x86_amd_ssb_disable();
1181		} else {
1182			x86_spec_ctrl_base |= SPEC_CTRL_SSBD;
1183			wrmsrl(MSR_IA32_SPEC_CTRL, x86_spec_ctrl_base);
1184		}
1185	}
1186
1187	return mode;
1188}
1189
1190static void ssb_select_mitigation(void)
1191{
1192	ssb_mode = __ssb_select_mitigation();
1193
1194	if (boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1195		pr_info("%s\n", ssb_strings[ssb_mode]);
1196}
1197
1198#undef pr_fmt
1199#define pr_fmt(fmt)     "Speculation prctl: " fmt
1200
1201static void task_update_spec_tif(struct task_struct *tsk)
1202{
1203	/* Force the update of the real TIF bits */
1204	set_tsk_thread_flag(tsk, TIF_SPEC_FORCE_UPDATE);
1205
1206	/*
1207	 * Immediately update the speculation control MSRs for the current
1208	 * task, but for a non-current task delay setting the CPU
1209	 * mitigation until it is scheduled next.
1210	 *
1211	 * This can only happen for SECCOMP mitigation. For PRCTL it's
1212	 * always the current task.
1213	 */
1214	if (tsk == current)
1215		speculation_ctrl_update_current();
1216}
1217
1218static int ssb_prctl_set(struct task_struct *task, unsigned long ctrl)
1219{
1220	if (ssb_mode != SPEC_STORE_BYPASS_PRCTL &&
1221	    ssb_mode != SPEC_STORE_BYPASS_SECCOMP)
1222		return -ENXIO;
1223
1224	switch (ctrl) {
1225	case PR_SPEC_ENABLE:
1226		/* If speculation is force disabled, enable is not allowed */
1227		if (task_spec_ssb_force_disable(task))
1228			return -EPERM;
1229		task_clear_spec_ssb_disable(task);
1230		task_clear_spec_ssb_noexec(task);
1231		task_update_spec_tif(task);
1232		break;
1233	case PR_SPEC_DISABLE:
1234		task_set_spec_ssb_disable(task);
1235		task_clear_spec_ssb_noexec(task);
1236		task_update_spec_tif(task);
1237		break;
1238	case PR_SPEC_FORCE_DISABLE:
1239		task_set_spec_ssb_disable(task);
1240		task_set_spec_ssb_force_disable(task);
1241		task_clear_spec_ssb_noexec(task);
1242		task_update_spec_tif(task);
1243		break;
1244	case PR_SPEC_DISABLE_NOEXEC:
1245		if (task_spec_ssb_force_disable(task))
1246			return -EPERM;
1247		task_set_spec_ssb_disable(task);
1248		task_set_spec_ssb_noexec(task);
1249		task_update_spec_tif(task);
1250		break;
1251	default:
1252		return -ERANGE;
1253	}
1254	return 0;
1255}
1256
1257static bool is_spec_ib_user_controlled(void)
1258{
1259	return spectre_v2_user_ibpb == SPECTRE_V2_USER_PRCTL ||
1260		spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP ||
1261		spectre_v2_user_stibp == SPECTRE_V2_USER_PRCTL ||
1262		spectre_v2_user_stibp == SPECTRE_V2_USER_SECCOMP;
1263}
1264
1265static int ib_prctl_set(struct task_struct *task, unsigned long ctrl)
1266{
1267	switch (ctrl) {
1268	case PR_SPEC_ENABLE:
1269		if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
1270		    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
1271			return 0;
1272
1273		/*
1274		 * With strict mode for both IBPB and STIBP, the instruction
1275		 * code paths avoid checking this task flag and instead,
1276		 * unconditionally run the instruction. However, STIBP and IBPB
1277		 * are independent and either can be set to conditionally
1278		 * enabled regardless of the mode of the other.
1279		 *
1280		 * If either is set to conditional, allow the task flag to be
1281		 * updated, unless it was force-disabled by a previous prctl
1282		 * call. Currently, this is possible on an AMD CPU which has the
1283		 * feature X86_FEATURE_AMD_STIBP_ALWAYS_ON. In this case, if the
1284		 * kernel is booted with 'spectre_v2_user=seccomp', then
1285		 * spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP and
1286		 * spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED.
1287		 */
1288		if (!is_spec_ib_user_controlled() ||
1289		    task_spec_ib_force_disable(task))
1290			return -EPERM;
1291
1292		task_clear_spec_ib_disable(task);
1293		task_update_spec_tif(task);
1294		break;
1295	case PR_SPEC_DISABLE:
1296	case PR_SPEC_FORCE_DISABLE:
1297		/*
1298		 * Indirect branch speculation is always allowed when
1299		 * mitigation is force disabled.
1300		 */
1301		if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
1302		    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
1303			return -EPERM;
1304
1305		if (!is_spec_ib_user_controlled())
1306			return 0;
1307
1308		task_set_spec_ib_disable(task);
1309		if (ctrl == PR_SPEC_FORCE_DISABLE)
1310			task_set_spec_ib_force_disable(task);
1311		task_update_spec_tif(task);
1312		break;
1313	default:
1314		return -ERANGE;
1315	}
1316	return 0;
1317}
1318
1319int arch_prctl_spec_ctrl_set(struct task_struct *task, unsigned long which,
1320			     unsigned long ctrl)
1321{
1322	switch (which) {
1323	case PR_SPEC_STORE_BYPASS:
1324		return ssb_prctl_set(task, ctrl);
1325	case PR_SPEC_INDIRECT_BRANCH:
1326		return ib_prctl_set(task, ctrl);
1327	default:
1328		return -ENODEV;
1329	}
1330}
1331
1332#ifdef CONFIG_SECCOMP
1333void arch_seccomp_spec_mitigate(struct task_struct *task)
1334{
1335	if (ssb_mode == SPEC_STORE_BYPASS_SECCOMP)
1336		ssb_prctl_set(task, PR_SPEC_FORCE_DISABLE);
1337	if (spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP ||
1338	    spectre_v2_user_stibp == SPECTRE_V2_USER_SECCOMP)
1339		ib_prctl_set(task, PR_SPEC_FORCE_DISABLE);
1340}
1341#endif
1342
1343static int ssb_prctl_get(struct task_struct *task)
1344{
1345	switch (ssb_mode) {
1346	case SPEC_STORE_BYPASS_DISABLE:
1347		return PR_SPEC_DISABLE;
1348	case SPEC_STORE_BYPASS_SECCOMP:
1349	case SPEC_STORE_BYPASS_PRCTL:
1350		if (task_spec_ssb_force_disable(task))
1351			return PR_SPEC_PRCTL | PR_SPEC_FORCE_DISABLE;
1352		if (task_spec_ssb_noexec(task))
1353			return PR_SPEC_PRCTL | PR_SPEC_DISABLE_NOEXEC;
1354		if (task_spec_ssb_disable(task))
1355			return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
1356		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
1357	default:
1358		if (boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1359			return PR_SPEC_ENABLE;
1360		return PR_SPEC_NOT_AFFECTED;
1361	}
1362}
1363
1364static int ib_prctl_get(struct task_struct *task)
1365{
1366	if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V2))
1367		return PR_SPEC_NOT_AFFECTED;
1368
1369	if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
1370	    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
1371		return PR_SPEC_ENABLE;
1372	else if (is_spec_ib_user_controlled()) {
1373		if (task_spec_ib_force_disable(task))
1374			return PR_SPEC_PRCTL | PR_SPEC_FORCE_DISABLE;
1375		if (task_spec_ib_disable(task))
1376			return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
1377		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
1378	} else if (spectre_v2_user_ibpb == SPECTRE_V2_USER_STRICT ||
1379	    spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT ||
1380	    spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED)
1381		return PR_SPEC_DISABLE;
1382	else
1383		return PR_SPEC_NOT_AFFECTED;
1384}
1385
1386int arch_prctl_spec_ctrl_get(struct task_struct *task, unsigned long which)
1387{
1388	switch (which) {
1389	case PR_SPEC_STORE_BYPASS:
1390		return ssb_prctl_get(task);
1391	case PR_SPEC_INDIRECT_BRANCH:
1392		return ib_prctl_get(task);
1393	default:
1394		return -ENODEV;
1395	}
1396}
1397
1398void x86_spec_ctrl_setup_ap(void)
1399{
1400	if (boot_cpu_has(X86_FEATURE_MSR_SPEC_CTRL))
1401		wrmsrl(MSR_IA32_SPEC_CTRL, x86_spec_ctrl_base);
1402
1403	if (ssb_mode == SPEC_STORE_BYPASS_DISABLE)
1404		x86_amd_ssb_disable();
1405}
1406
1407bool itlb_multihit_kvm_mitigation;
1408EXPORT_SYMBOL_GPL(itlb_multihit_kvm_mitigation);
1409
1410#undef pr_fmt
1411#define pr_fmt(fmt)	"L1TF: " fmt
1412
1413/* Default mitigation for L1TF-affected CPUs */
1414enum l1tf_mitigations l1tf_mitigation __ro_after_init = L1TF_MITIGATION_FLUSH;
1415#if IS_ENABLED(CONFIG_KVM_INTEL)
1416EXPORT_SYMBOL_GPL(l1tf_mitigation);
1417#endif
1418enum vmx_l1d_flush_state l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
1419EXPORT_SYMBOL_GPL(l1tf_vmx_mitigation);
1420
1421/*
1422 * These CPUs all support 44bits physical address space internally in the
1423 * cache but CPUID can report a smaller number of physical address bits.
1424 *
1425 * The L1TF mitigation uses the top most address bit for the inversion of
1426 * non present PTEs. When the installed memory reaches into the top most
1427 * address bit due to memory holes, which has been observed on machines
1428 * which report 36bits physical address bits and have 32G RAM installed,
1429 * then the mitigation range check in l1tf_select_mitigation() triggers.
1430 * This is a false positive because the mitigation is still possible due to
1431 * the fact that the cache uses 44bit internally. Use the cache bits
1432 * instead of the reported physical bits and adjust them on the affected
1433 * machines to 44bit if the reported bits are less than 44.
1434 */
1435static void override_cache_bits(struct cpuinfo_x86 *c)
1436{
1437	if (c->x86 != 6)
1438		return;
1439
1440	switch (c->x86_model) {
1441	case INTEL_FAM6_NEHALEM:
1442	case INTEL_FAM6_WESTMERE:
1443	case INTEL_FAM6_SANDYBRIDGE:
1444	case INTEL_FAM6_IVYBRIDGE:
1445	case INTEL_FAM6_HASWELL:
1446	case INTEL_FAM6_HASWELL_L:
1447	case INTEL_FAM6_HASWELL_G:
1448	case INTEL_FAM6_BROADWELL:
1449	case INTEL_FAM6_BROADWELL_G:
1450	case INTEL_FAM6_SKYLAKE_L:
1451	case INTEL_FAM6_SKYLAKE:
1452	case INTEL_FAM6_KABYLAKE_L:
1453	case INTEL_FAM6_KABYLAKE:
1454		if (c->x86_cache_bits < 44)
1455			c->x86_cache_bits = 44;
1456		break;
1457	}
1458}
1459
1460static void __init l1tf_select_mitigation(void)
1461{
1462	u64 half_pa;
1463
1464	if (!boot_cpu_has_bug(X86_BUG_L1TF))
1465		return;
1466
1467	if (cpu_mitigations_off())
1468		l1tf_mitigation = L1TF_MITIGATION_OFF;
1469	else if (cpu_mitigations_auto_nosmt())
1470		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOSMT;
1471
1472	override_cache_bits(&boot_cpu_data);
1473
1474	switch (l1tf_mitigation) {
1475	case L1TF_MITIGATION_OFF:
1476	case L1TF_MITIGATION_FLUSH_NOWARN:
1477	case L1TF_MITIGATION_FLUSH:
1478		break;
1479	case L1TF_MITIGATION_FLUSH_NOSMT:
1480	case L1TF_MITIGATION_FULL:
1481		cpu_smt_disable(false);
1482		break;
1483	case L1TF_MITIGATION_FULL_FORCE:
1484		cpu_smt_disable(true);
1485		break;
1486	}
1487
1488#if CONFIG_PGTABLE_LEVELS == 2
1489	pr_warn("Kernel not compiled for PAE. No mitigation for L1TF\n");
1490	return;
1491#endif
1492
1493	half_pa = (u64)l1tf_pfn_limit() << PAGE_SHIFT;
1494	if (l1tf_mitigation != L1TF_MITIGATION_OFF &&
1495			e820__mapped_any(half_pa, ULLONG_MAX - half_pa, E820_TYPE_RAM)) {
1496		pr_warn("System has more than MAX_PA/2 memory. L1TF mitigation not effective.\n");
1497		pr_info("You may make it effective by booting the kernel with mem=%llu parameter.\n",
1498				half_pa);
1499		pr_info("However, doing so will make a part of your RAM unusable.\n");
1500		pr_info("Reading https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/l1tf.html might help you decide.\n");
1501		return;
1502	}
1503
1504	setup_force_cpu_cap(X86_FEATURE_L1TF_PTEINV);
1505}
1506
1507static int __init l1tf_cmdline(char *str)
1508{
1509	if (!boot_cpu_has_bug(X86_BUG_L1TF))
1510		return 0;
1511
1512	if (!str)
1513		return -EINVAL;
1514
1515	if (!strcmp(str, "off"))
1516		l1tf_mitigation = L1TF_MITIGATION_OFF;
1517	else if (!strcmp(str, "flush,nowarn"))
1518		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOWARN;
1519	else if (!strcmp(str, "flush"))
1520		l1tf_mitigation = L1TF_MITIGATION_FLUSH;
1521	else if (!strcmp(str, "flush,nosmt"))
1522		l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOSMT;
1523	else if (!strcmp(str, "full"))
1524		l1tf_mitigation = L1TF_MITIGATION_FULL;
1525	else if (!strcmp(str, "full,force"))
1526		l1tf_mitigation = L1TF_MITIGATION_FULL_FORCE;
1527
1528	return 0;
1529}
1530early_param("l1tf", l1tf_cmdline);
1531
1532#undef pr_fmt
1533#define pr_fmt(fmt) fmt
1534
1535#ifdef CONFIG_SYSFS
1536
1537#define L1TF_DEFAULT_MSG "Mitigation: PTE Inversion"
1538
1539#if IS_ENABLED(CONFIG_KVM_INTEL)
1540static const char * const l1tf_vmx_states[] = {
1541	[VMENTER_L1D_FLUSH_AUTO]		= "auto",
1542	[VMENTER_L1D_FLUSH_NEVER]		= "vulnerable",
1543	[VMENTER_L1D_FLUSH_COND]		= "conditional cache flushes",
1544	[VMENTER_L1D_FLUSH_ALWAYS]		= "cache flushes",
1545	[VMENTER_L1D_FLUSH_EPT_DISABLED]	= "EPT disabled",
1546	[VMENTER_L1D_FLUSH_NOT_REQUIRED]	= "flush not necessary"
1547};
1548
1549static ssize_t l1tf_show_state(char *buf)
1550{
1551	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO)
1552		return sprintf(buf, "%s\n", L1TF_DEFAULT_MSG);
1553
1554	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_EPT_DISABLED ||
1555	    (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER &&
1556	     sched_smt_active())) {
1557		return sprintf(buf, "%s; VMX: %s\n", L1TF_DEFAULT_MSG,
1558			       l1tf_vmx_states[l1tf_vmx_mitigation]);
1559	}
1560
1561	return sprintf(buf, "%s; VMX: %s, SMT %s\n", L1TF_DEFAULT_MSG,
1562		       l1tf_vmx_states[l1tf_vmx_mitigation],
1563		       sched_smt_active() ? "vulnerable" : "disabled");
1564}
1565
1566static ssize_t itlb_multihit_show_state(char *buf)
1567{
1568	if (!boot_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
1569	    !boot_cpu_has(X86_FEATURE_VMX))
1570		return sprintf(buf, "KVM: Mitigation: VMX unsupported\n");
1571	else if (!(cr4_read_shadow() & X86_CR4_VMXE))
1572		return sprintf(buf, "KVM: Mitigation: VMX disabled\n");
1573	else if (itlb_multihit_kvm_mitigation)
1574		return sprintf(buf, "KVM: Mitigation: Split huge pages\n");
1575	else
1576		return sprintf(buf, "KVM: Vulnerable\n");
1577}
1578#else
1579static ssize_t l1tf_show_state(char *buf)
1580{
1581	return sprintf(buf, "%s\n", L1TF_DEFAULT_MSG);
1582}
1583
1584static ssize_t itlb_multihit_show_state(char *buf)
1585{
1586	return sprintf(buf, "Processor vulnerable\n");
1587}
1588#endif
1589
1590static ssize_t mds_show_state(char *buf)
1591{
1592	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
1593		return sprintf(buf, "%s; SMT Host state unknown\n",
1594			       mds_strings[mds_mitigation]);
1595	}
1596
1597	if (boot_cpu_has(X86_BUG_MSBDS_ONLY)) {
1598		return sprintf(buf, "%s; SMT %s\n", mds_strings[mds_mitigation],
1599			       (mds_mitigation == MDS_MITIGATION_OFF ? "vulnerable" :
1600			        sched_smt_active() ? "mitigated" : "disabled"));
1601	}
1602
1603	return sprintf(buf, "%s; SMT %s\n", mds_strings[mds_mitigation],
1604		       sched_smt_active() ? "vulnerable" : "disabled");
1605}
1606
1607static ssize_t tsx_async_abort_show_state(char *buf)
1608{
1609	if ((taa_mitigation == TAA_MITIGATION_TSX_DISABLED) ||
1610	    (taa_mitigation == TAA_MITIGATION_OFF))
1611		return sprintf(buf, "%s\n", taa_strings[taa_mitigation]);
1612
1613	if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
1614		return sprintf(buf, "%s; SMT Host state unknown\n",
1615			       taa_strings[taa_mitigation]);
1616	}
1617
1618	return sprintf(buf, "%s; SMT %s\n", taa_strings[taa_mitigation],
1619		       sched_smt_active() ? "vulnerable" : "disabled");
1620}
1621
1622static char *stibp_state(void)
1623{
1624	if (spectre_v2_enabled == SPECTRE_V2_IBRS_ENHANCED)
1625		return "";
1626
1627	switch (spectre_v2_user_stibp) {
1628	case SPECTRE_V2_USER_NONE:
1629		return ", STIBP: disabled";
1630	case SPECTRE_V2_USER_STRICT:
1631		return ", STIBP: forced";
1632	case SPECTRE_V2_USER_STRICT_PREFERRED:
1633		return ", STIBP: always-on";
1634	case SPECTRE_V2_USER_PRCTL:
1635	case SPECTRE_V2_USER_SECCOMP:
1636		if (static_key_enabled(&switch_to_cond_stibp))
1637			return ", STIBP: conditional";
1638	}
1639	return "";
1640}
1641
1642static char *ibpb_state(void)
1643{
1644	if (boot_cpu_has(X86_FEATURE_IBPB)) {
1645		if (static_key_enabled(&switch_mm_always_ibpb))
1646			return ", IBPB: always-on";
1647		if (static_key_enabled(&switch_mm_cond_ibpb))
1648			return ", IBPB: conditional";
1649		return ", IBPB: disabled";
1650	}
1651	return "";
1652}
1653
1654static ssize_t srbds_show_state(char *buf)
1655{
1656	return sprintf(buf, "%s\n", srbds_strings[srbds_mitigation]);
1657}
1658
1659static ssize_t cpu_show_common(struct device *dev, struct device_attribute *attr,
1660			       char *buf, unsigned int bug)
1661{
1662	if (!boot_cpu_has_bug(bug))
1663		return sprintf(buf, "Not affected\n");
1664
1665	switch (bug) {
1666	case X86_BUG_CPU_MELTDOWN:
1667		if (boot_cpu_has(X86_FEATURE_PTI))
1668			return sprintf(buf, "Mitigation: PTI\n");
1669
1670		if (hypervisor_is_type(X86_HYPER_XEN_PV))
1671			return sprintf(buf, "Unknown (XEN PV detected, hypervisor mitigation required)\n");
1672
1673		break;
1674
1675	case X86_BUG_SPECTRE_V1:
1676		return sprintf(buf, "%s\n", spectre_v1_strings[spectre_v1_mitigation]);
1677
1678	case X86_BUG_SPECTRE_V2:
1679		return sprintf(buf, "%s%s%s%s%s%s\n", spectre_v2_strings[spectre_v2_enabled],
1680			       ibpb_state(),
1681			       boot_cpu_has(X86_FEATURE_USE_IBRS_FW) ? ", IBRS_FW" : "",
1682			       stibp_state(),
1683			       boot_cpu_has(X86_FEATURE_RSB_CTXSW) ? ", RSB filling" : "",
1684			       spectre_v2_module_string());
1685
1686	case X86_BUG_SPEC_STORE_BYPASS:
1687		return sprintf(buf, "%s\n", ssb_strings[ssb_mode]);
1688
1689	case X86_BUG_L1TF:
1690		if (boot_cpu_has(X86_FEATURE_L1TF_PTEINV))
1691			return l1tf_show_state(buf);
1692		break;
1693
1694	case X86_BUG_MDS:
1695		return mds_show_state(buf);
1696
1697	case X86_BUG_TAA:
1698		return tsx_async_abort_show_state(buf);
1699
1700	case X86_BUG_ITLB_MULTIHIT:
1701		return itlb_multihit_show_state(buf);
1702
1703	case X86_BUG_SRBDS:
1704		return srbds_show_state(buf);
1705
1706	default:
1707		break;
1708	}
1709
1710	return sprintf(buf, "Vulnerable\n");
1711}
1712
1713ssize_t cpu_show_meltdown(struct device *dev, struct device_attribute *attr, char *buf)
1714{
1715	return cpu_show_common(dev, attr, buf, X86_BUG_CPU_MELTDOWN);
1716}
1717
1718ssize_t cpu_show_spectre_v1(struct device *dev, struct device_attribute *attr, char *buf)
1719{
1720	return cpu_show_common(dev, attr, buf, X86_BUG_SPECTRE_V1);
1721}
1722
1723ssize_t cpu_show_spectre_v2(struct device *dev, struct device_attribute *attr, char *buf)
1724{
1725	return cpu_show_common(dev, attr, buf, X86_BUG_SPECTRE_V2);
1726}
1727
1728ssize_t cpu_show_spec_store_bypass(struct device *dev, struct device_attribute *attr, char *buf)
1729{
1730	return cpu_show_common(dev, attr, buf, X86_BUG_SPEC_STORE_BYPASS);
1731}
1732
1733ssize_t cpu_show_l1tf(struct device *dev, struct device_attribute *attr, char *buf)
1734{
1735	return cpu_show_common(dev, attr, buf, X86_BUG_L1TF);
1736}
1737
1738ssize_t cpu_show_mds(struct device *dev, struct device_attribute *attr, char *buf)
1739{
1740	return cpu_show_common(dev, attr, buf, X86_BUG_MDS);
1741}
1742
1743ssize_t cpu_show_tsx_async_abort(struct device *dev, struct device_attribute *attr, char *buf)
1744{
1745	return cpu_show_common(dev, attr, buf, X86_BUG_TAA);
1746}
1747
1748ssize_t cpu_show_itlb_multihit(struct device *dev, struct device_attribute *attr, char *buf)
1749{
1750	return cpu_show_common(dev, attr, buf, X86_BUG_ITLB_MULTIHIT);
1751}
1752
1753ssize_t cpu_show_srbds(struct device *dev, struct device_attribute *attr, char *buf)
1754{
1755	return cpu_show_common(dev, attr, buf, X86_BUG_SRBDS);
1756}
1757#endif