]>
Commit | Line | Data |
---|---|---|
1 | // SPDX-License-Identifier: GPL-2.0-only | |
2 | /* cpu_feature_enabled() cannot be used this early */ | |
3 | #define USE_EARLY_PGTABLE_L5 | |
4 | ||
5 | #include <linux/memblock.h> | |
6 | #include <linux/linkage.h> | |
7 | #include <linux/bitops.h> | |
8 | #include <linux/kernel.h> | |
9 | #include <linux/export.h> | |
10 | #include <linux/percpu.h> | |
11 | #include <linux/string.h> | |
12 | #include <linux/ctype.h> | |
13 | #include <linux/delay.h> | |
14 | #include <linux/sched/mm.h> | |
15 | #include <linux/sched/clock.h> | |
16 | #include <linux/sched/task.h> | |
17 | #include <linux/sched/smt.h> | |
18 | #include <linux/init.h> | |
19 | #include <linux/kprobes.h> | |
20 | #include <linux/kgdb.h> | |
21 | #include <linux/smp.h> | |
22 | #include <linux/io.h> | |
23 | #include <linux/syscore_ops.h> | |
24 | #include <linux/pgtable.h> | |
25 | ||
26 | #include <asm/cmdline.h> | |
27 | #include <asm/stackprotector.h> | |
28 | #include <asm/perf_event.h> | |
29 | #include <asm/mmu_context.h> | |
30 | #include <asm/doublefault.h> | |
31 | #include <asm/archrandom.h> | |
32 | #include <asm/hypervisor.h> | |
33 | #include <asm/processor.h> | |
34 | #include <asm/tlbflush.h> | |
35 | #include <asm/debugreg.h> | |
36 | #include <asm/sections.h> | |
37 | #include <asm/vsyscall.h> | |
38 | #include <linux/topology.h> | |
39 | #include <linux/cpumask.h> | |
40 | #include <linux/atomic.h> | |
41 | #include <asm/proto.h> | |
42 | #include <asm/setup.h> | |
43 | #include <asm/apic.h> | |
44 | #include <asm/desc.h> | |
45 | #include <asm/fpu/api.h> | |
46 | #include <asm/mtrr.h> | |
47 | #include <asm/hwcap2.h> | |
48 | #include <linux/numa.h> | |
49 | #include <asm/numa.h> | |
50 | #include <asm/asm.h> | |
51 | #include <asm/bugs.h> | |
52 | #include <asm/cpu.h> | |
53 | #include <asm/mce.h> | |
54 | #include <asm/msr.h> | |
55 | #include <asm/memtype.h> | |
56 | #include <asm/microcode.h> | |
57 | #include <asm/microcode_intel.h> | |
58 | #include <asm/intel-family.h> | |
59 | #include <asm/cpu_device_id.h> | |
60 | #include <asm/uv/uv.h> | |
61 | #include <asm/sigframe.h> | |
62 | ||
63 | #include "cpu.h" | |
64 | ||
65 | u32 elf_hwcap2 __read_mostly; | |
66 | ||
67 | /* all of these masks are initialized in setup_cpu_local_masks() */ | |
68 | cpumask_var_t cpu_initialized_mask; | |
69 | cpumask_var_t cpu_callout_mask; | |
70 | cpumask_var_t cpu_callin_mask; | |
71 | ||
72 | /* representing cpus for which sibling maps can be computed */ | |
73 | cpumask_var_t cpu_sibling_setup_mask; | |
74 | ||
75 | /* Number of siblings per CPU package */ | |
76 | int smp_num_siblings = 1; | |
77 | EXPORT_SYMBOL(smp_num_siblings); | |
78 | ||
79 | /* Last level cache ID of each logical CPU */ | |
80 | DEFINE_PER_CPU_READ_MOSTLY(u16, cpu_llc_id) = BAD_APICID; | |
81 | ||
82 | u16 get_llc_id(unsigned int cpu) | |
83 | { | |
84 | return per_cpu(cpu_llc_id, cpu); | |
85 | } | |
86 | EXPORT_SYMBOL_GPL(get_llc_id); | |
87 | ||
88 | /* correctly size the local cpu masks */ | |
89 | void __init setup_cpu_local_masks(void) | |
90 | { | |
91 | alloc_bootmem_cpumask_var(&cpu_initialized_mask); | |
92 | alloc_bootmem_cpumask_var(&cpu_callin_mask); | |
93 | alloc_bootmem_cpumask_var(&cpu_callout_mask); | |
94 | alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask); | |
95 | } | |
96 | ||
97 | static void default_init(struct cpuinfo_x86 *c) | |
98 | { | |
99 | #ifdef CONFIG_X86_64 | |
100 | cpu_detect_cache_sizes(c); | |
101 | #else | |
102 | /* Not much we can do here... */ | |
103 | /* Check if at least it has cpuid */ | |
104 | if (c->cpuid_level == -1) { | |
105 | /* No cpuid. It must be an ancient CPU */ | |
106 | if (c->x86 == 4) | |
107 | strcpy(c->x86_model_id, "486"); | |
108 | else if (c->x86 == 3) | |
109 | strcpy(c->x86_model_id, "386"); | |
110 | } | |
111 | #endif | |
112 | } | |
113 | ||
114 | static const struct cpu_dev default_cpu = { | |
115 | .c_init = default_init, | |
116 | .c_vendor = "Unknown", | |
117 | .c_x86_vendor = X86_VENDOR_UNKNOWN, | |
118 | }; | |
119 | ||
120 | static const struct cpu_dev *this_cpu = &default_cpu; | |
121 | ||
122 | DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = { | |
123 | #ifdef CONFIG_X86_64 | |
124 | /* | |
125 | * We need valid kernel segments for data and code in long mode too | |
126 | * IRET will check the segment types kkeil 2000/10/28 | |
127 | * Also sysret mandates a special GDT layout | |
128 | * | |
129 | * TLS descriptors are currently at a different place compared to i386. | |
130 | * Hopefully nobody expects them at a fixed place (Wine?) | |
131 | */ | |
132 | [GDT_ENTRY_KERNEL32_CS] = GDT_ENTRY_INIT(0xc09b, 0, 0xfffff), | |
133 | [GDT_ENTRY_KERNEL_CS] = GDT_ENTRY_INIT(0xa09b, 0, 0xfffff), | |
134 | [GDT_ENTRY_KERNEL_DS] = GDT_ENTRY_INIT(0xc093, 0, 0xfffff), | |
135 | [GDT_ENTRY_DEFAULT_USER32_CS] = GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff), | |
136 | [GDT_ENTRY_DEFAULT_USER_DS] = GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff), | |
137 | [GDT_ENTRY_DEFAULT_USER_CS] = GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff), | |
138 | #else | |
139 | [GDT_ENTRY_KERNEL_CS] = GDT_ENTRY_INIT(0xc09a, 0, 0xfffff), | |
140 | [GDT_ENTRY_KERNEL_DS] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff), | |
141 | [GDT_ENTRY_DEFAULT_USER_CS] = GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff), | |
142 | [GDT_ENTRY_DEFAULT_USER_DS] = GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff), | |
143 | /* | |
144 | * Segments used for calling PnP BIOS have byte granularity. | |
145 | * They code segments and data segments have fixed 64k limits, | |
146 | * the transfer segment sizes are set at run time. | |
147 | */ | |
148 | /* 32-bit code */ | |
149 | [GDT_ENTRY_PNPBIOS_CS32] = GDT_ENTRY_INIT(0x409a, 0, 0xffff), | |
150 | /* 16-bit code */ | |
151 | [GDT_ENTRY_PNPBIOS_CS16] = GDT_ENTRY_INIT(0x009a, 0, 0xffff), | |
152 | /* 16-bit data */ | |
153 | [GDT_ENTRY_PNPBIOS_DS] = GDT_ENTRY_INIT(0x0092, 0, 0xffff), | |
154 | /* 16-bit data */ | |
155 | [GDT_ENTRY_PNPBIOS_TS1] = GDT_ENTRY_INIT(0x0092, 0, 0), | |
156 | /* 16-bit data */ | |
157 | [GDT_ENTRY_PNPBIOS_TS2] = GDT_ENTRY_INIT(0x0092, 0, 0), | |
158 | /* | |
159 | * The APM segments have byte granularity and their bases | |
160 | * are set at run time. All have 64k limits. | |
161 | */ | |
162 | /* 32-bit code */ | |
163 | [GDT_ENTRY_APMBIOS_BASE] = GDT_ENTRY_INIT(0x409a, 0, 0xffff), | |
164 | /* 16-bit code */ | |
165 | [GDT_ENTRY_APMBIOS_BASE+1] = GDT_ENTRY_INIT(0x009a, 0, 0xffff), | |
166 | /* data */ | |
167 | [GDT_ENTRY_APMBIOS_BASE+2] = GDT_ENTRY_INIT(0x4092, 0, 0xffff), | |
168 | ||
169 | [GDT_ENTRY_ESPFIX_SS] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff), | |
170 | [GDT_ENTRY_PERCPU] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff), | |
171 | #endif | |
172 | } }; | |
173 | EXPORT_PER_CPU_SYMBOL_GPL(gdt_page); | |
174 | ||
175 | #ifdef CONFIG_X86_64 | |
176 | static int __init x86_nopcid_setup(char *s) | |
177 | { | |
178 | /* nopcid doesn't accept parameters */ | |
179 | if (s) | |
180 | return -EINVAL; | |
181 | ||
182 | /* do not emit a message if the feature is not present */ | |
183 | if (!boot_cpu_has(X86_FEATURE_PCID)) | |
184 | return 0; | |
185 | ||
186 | setup_clear_cpu_cap(X86_FEATURE_PCID); | |
187 | pr_info("nopcid: PCID feature disabled\n"); | |
188 | return 0; | |
189 | } | |
190 | early_param("nopcid", x86_nopcid_setup); | |
191 | #endif | |
192 | ||
193 | static int __init x86_noinvpcid_setup(char *s) | |
194 | { | |
195 | /* noinvpcid doesn't accept parameters */ | |
196 | if (s) | |
197 | return -EINVAL; | |
198 | ||
199 | /* do not emit a message if the feature is not present */ | |
200 | if (!boot_cpu_has(X86_FEATURE_INVPCID)) | |
201 | return 0; | |
202 | ||
203 | setup_clear_cpu_cap(X86_FEATURE_INVPCID); | |
204 | pr_info("noinvpcid: INVPCID feature disabled\n"); | |
205 | return 0; | |
206 | } | |
207 | early_param("noinvpcid", x86_noinvpcid_setup); | |
208 | ||
209 | #ifdef CONFIG_X86_32 | |
210 | static int cachesize_override = -1; | |
211 | static int disable_x86_serial_nr = 1; | |
212 | ||
213 | static int __init cachesize_setup(char *str) | |
214 | { | |
215 | get_option(&str, &cachesize_override); | |
216 | return 1; | |
217 | } | |
218 | __setup("cachesize=", cachesize_setup); | |
219 | ||
220 | static int __init x86_sep_setup(char *s) | |
221 | { | |
222 | setup_clear_cpu_cap(X86_FEATURE_SEP); | |
223 | return 1; | |
224 | } | |
225 | __setup("nosep", x86_sep_setup); | |
226 | ||
227 | /* Standard macro to see if a specific flag is changeable */ | |
228 | static inline int flag_is_changeable_p(u32 flag) | |
229 | { | |
230 | u32 f1, f2; | |
231 | ||
232 | /* | |
233 | * Cyrix and IDT cpus allow disabling of CPUID | |
234 | * so the code below may return different results | |
235 | * when it is executed before and after enabling | |
236 | * the CPUID. Add "volatile" to not allow gcc to | |
237 | * optimize the subsequent calls to this function. | |
238 | */ | |
239 | asm volatile ("pushfl \n\t" | |
240 | "pushfl \n\t" | |
241 | "popl %0 \n\t" | |
242 | "movl %0, %1 \n\t" | |
243 | "xorl %2, %0 \n\t" | |
244 | "pushl %0 \n\t" | |
245 | "popfl \n\t" | |
246 | "pushfl \n\t" | |
247 | "popl %0 \n\t" | |
248 | "popfl \n\t" | |
249 | ||
250 | : "=&r" (f1), "=&r" (f2) | |
251 | : "ir" (flag)); | |
252 | ||
253 | return ((f1^f2) & flag) != 0; | |
254 | } | |
255 | ||
256 | /* Probe for the CPUID instruction */ | |
257 | int have_cpuid_p(void) | |
258 | { | |
259 | return flag_is_changeable_p(X86_EFLAGS_ID); | |
260 | } | |
261 | ||
262 | static void squash_the_stupid_serial_number(struct cpuinfo_x86 *c) | |
263 | { | |
264 | unsigned long lo, hi; | |
265 | ||
266 | if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr) | |
267 | return; | |
268 | ||
269 | /* Disable processor serial number: */ | |
270 | ||
271 | rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi); | |
272 | lo |= 0x200000; | |
273 | wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi); | |
274 | ||
275 | pr_notice("CPU serial number disabled.\n"); | |
276 | clear_cpu_cap(c, X86_FEATURE_PN); | |
277 | ||
278 | /* Disabling the serial number may affect the cpuid level */ | |
279 | c->cpuid_level = cpuid_eax(0); | |
280 | } | |
281 | ||
282 | static int __init x86_serial_nr_setup(char *s) | |
283 | { | |
284 | disable_x86_serial_nr = 0; | |
285 | return 1; | |
286 | } | |
287 | __setup("serialnumber", x86_serial_nr_setup); | |
288 | #else | |
289 | static inline int flag_is_changeable_p(u32 flag) | |
290 | { | |
291 | return 1; | |
292 | } | |
293 | static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c) | |
294 | { | |
295 | } | |
296 | #endif | |
297 | ||
298 | static __init int setup_disable_smep(char *arg) | |
299 | { | |
300 | setup_clear_cpu_cap(X86_FEATURE_SMEP); | |
301 | return 1; | |
302 | } | |
303 | __setup("nosmep", setup_disable_smep); | |
304 | ||
305 | static __always_inline void setup_smep(struct cpuinfo_x86 *c) | |
306 | { | |
307 | if (cpu_has(c, X86_FEATURE_SMEP)) | |
308 | cr4_set_bits(X86_CR4_SMEP); | |
309 | } | |
310 | ||
311 | static __init int setup_disable_smap(char *arg) | |
312 | { | |
313 | setup_clear_cpu_cap(X86_FEATURE_SMAP); | |
314 | return 1; | |
315 | } | |
316 | __setup("nosmap", setup_disable_smap); | |
317 | ||
318 | static __always_inline void setup_smap(struct cpuinfo_x86 *c) | |
319 | { | |
320 | unsigned long eflags = native_save_fl(); | |
321 | ||
322 | /* This should have been cleared long ago */ | |
323 | BUG_ON(eflags & X86_EFLAGS_AC); | |
324 | ||
325 | if (cpu_has(c, X86_FEATURE_SMAP)) { | |
326 | #ifdef CONFIG_X86_SMAP | |
327 | cr4_set_bits(X86_CR4_SMAP); | |
328 | #else | |
329 | clear_cpu_cap(c, X86_FEATURE_SMAP); | |
330 | cr4_clear_bits(X86_CR4_SMAP); | |
331 | #endif | |
332 | } | |
333 | } | |
334 | ||
335 | static __always_inline void setup_umip(struct cpuinfo_x86 *c) | |
336 | { | |
337 | /* Check the boot processor, plus build option for UMIP. */ | |
338 | if (!cpu_feature_enabled(X86_FEATURE_UMIP)) | |
339 | goto out; | |
340 | ||
341 | /* Check the current processor's cpuid bits. */ | |
342 | if (!cpu_has(c, X86_FEATURE_UMIP)) | |
343 | goto out; | |
344 | ||
345 | cr4_set_bits(X86_CR4_UMIP); | |
346 | ||
347 | pr_info_once("x86/cpu: User Mode Instruction Prevention (UMIP) activated\n"); | |
348 | ||
349 | return; | |
350 | ||
351 | out: | |
352 | /* | |
353 | * Make sure UMIP is disabled in case it was enabled in a | |
354 | * previous boot (e.g., via kexec). | |
355 | */ | |
356 | cr4_clear_bits(X86_CR4_UMIP); | |
357 | } | |
358 | ||
359 | /* These bits should not change their value after CPU init is finished. */ | |
360 | static const unsigned long cr4_pinned_mask = | |
361 | X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_UMIP | X86_CR4_FSGSBASE; | |
362 | static DEFINE_STATIC_KEY_FALSE_RO(cr_pinning); | |
363 | static unsigned long cr4_pinned_bits __ro_after_init; | |
364 | ||
365 | void native_write_cr0(unsigned long val) | |
366 | { | |
367 | unsigned long bits_missing = 0; | |
368 | ||
369 | set_register: | |
370 | asm volatile("mov %0,%%cr0": "+r" (val) : : "memory"); | |
371 | ||
372 | if (static_branch_likely(&cr_pinning)) { | |
373 | if (unlikely((val & X86_CR0_WP) != X86_CR0_WP)) { | |
374 | bits_missing = X86_CR0_WP; | |
375 | val |= bits_missing; | |
376 | goto set_register; | |
377 | } | |
378 | /* Warn after we've set the missing bits. */ | |
379 | WARN_ONCE(bits_missing, "CR0 WP bit went missing!?\n"); | |
380 | } | |
381 | } | |
382 | EXPORT_SYMBOL(native_write_cr0); | |
383 | ||
384 | void native_write_cr4(unsigned long val) | |
385 | { | |
386 | unsigned long bits_changed = 0; | |
387 | ||
388 | set_register: | |
389 | asm volatile("mov %0,%%cr4": "+r" (val) : : "memory"); | |
390 | ||
391 | if (static_branch_likely(&cr_pinning)) { | |
392 | if (unlikely((val & cr4_pinned_mask) != cr4_pinned_bits)) { | |
393 | bits_changed = (val & cr4_pinned_mask) ^ cr4_pinned_bits; | |
394 | val = (val & ~cr4_pinned_mask) | cr4_pinned_bits; | |
395 | goto set_register; | |
396 | } | |
397 | /* Warn after we've corrected the changed bits. */ | |
398 | WARN_ONCE(bits_changed, "pinned CR4 bits changed: 0x%lx!?\n", | |
399 | bits_changed); | |
400 | } | |
401 | } | |
402 | #if IS_MODULE(CONFIG_LKDTM) | |
403 | EXPORT_SYMBOL_GPL(native_write_cr4); | |
404 | #endif | |
405 | ||
406 | void cr4_update_irqsoff(unsigned long set, unsigned long clear) | |
407 | { | |
408 | unsigned long newval, cr4 = this_cpu_read(cpu_tlbstate.cr4); | |
409 | ||
410 | lockdep_assert_irqs_disabled(); | |
411 | ||
412 | newval = (cr4 & ~clear) | set; | |
413 | if (newval != cr4) { | |
414 | this_cpu_write(cpu_tlbstate.cr4, newval); | |
415 | __write_cr4(newval); | |
416 | } | |
417 | } | |
418 | EXPORT_SYMBOL(cr4_update_irqsoff); | |
419 | ||
420 | /* Read the CR4 shadow. */ | |
421 | unsigned long cr4_read_shadow(void) | |
422 | { | |
423 | return this_cpu_read(cpu_tlbstate.cr4); | |
424 | } | |
425 | EXPORT_SYMBOL_GPL(cr4_read_shadow); | |
426 | ||
427 | void cr4_init(void) | |
428 | { | |
429 | unsigned long cr4 = __read_cr4(); | |
430 | ||
431 | if (boot_cpu_has(X86_FEATURE_PCID)) | |
432 | cr4 |= X86_CR4_PCIDE; | |
433 | if (static_branch_likely(&cr_pinning)) | |
434 | cr4 = (cr4 & ~cr4_pinned_mask) | cr4_pinned_bits; | |
435 | ||
436 | __write_cr4(cr4); | |
437 | ||
438 | /* Initialize cr4 shadow for this CPU. */ | |
439 | this_cpu_write(cpu_tlbstate.cr4, cr4); | |
440 | } | |
441 | ||
442 | /* | |
443 | * Once CPU feature detection is finished (and boot params have been | |
444 | * parsed), record any of the sensitive CR bits that are set, and | |
445 | * enable CR pinning. | |
446 | */ | |
447 | static void __init setup_cr_pinning(void) | |
448 | { | |
449 | cr4_pinned_bits = this_cpu_read(cpu_tlbstate.cr4) & cr4_pinned_mask; | |
450 | static_key_enable(&cr_pinning.key); | |
451 | } | |
452 | ||
453 | static __init int x86_nofsgsbase_setup(char *arg) | |
454 | { | |
455 | /* Require an exact match without trailing characters. */ | |
456 | if (strlen(arg)) | |
457 | return 0; | |
458 | ||
459 | /* Do not emit a message if the feature is not present. */ | |
460 | if (!boot_cpu_has(X86_FEATURE_FSGSBASE)) | |
461 | return 1; | |
462 | ||
463 | setup_clear_cpu_cap(X86_FEATURE_FSGSBASE); | |
464 | pr_info("FSGSBASE disabled via kernel command line\n"); | |
465 | return 1; | |
466 | } | |
467 | __setup("nofsgsbase", x86_nofsgsbase_setup); | |
468 | ||
469 | /* | |
470 | * Protection Keys are not available in 32-bit mode. | |
471 | */ | |
472 | static bool pku_disabled; | |
473 | ||
474 | static __always_inline void setup_pku(struct cpuinfo_x86 *c) | |
475 | { | |
476 | if (c == &boot_cpu_data) { | |
477 | if (pku_disabled || !cpu_feature_enabled(X86_FEATURE_PKU)) | |
478 | return; | |
479 | /* | |
480 | * Setting CR4.PKE will cause the X86_FEATURE_OSPKE cpuid | |
481 | * bit to be set. Enforce it. | |
482 | */ | |
483 | setup_force_cpu_cap(X86_FEATURE_OSPKE); | |
484 | ||
485 | } else if (!cpu_feature_enabled(X86_FEATURE_OSPKE)) { | |
486 | return; | |
487 | } | |
488 | ||
489 | cr4_set_bits(X86_CR4_PKE); | |
490 | /* Load the default PKRU value */ | |
491 | pkru_write_default(); | |
492 | } | |
493 | ||
494 | #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS | |
495 | static __init int setup_disable_pku(char *arg) | |
496 | { | |
497 | /* | |
498 | * Do not clear the X86_FEATURE_PKU bit. All of the | |
499 | * runtime checks are against OSPKE so clearing the | |
500 | * bit does nothing. | |
501 | * | |
502 | * This way, we will see "pku" in cpuinfo, but not | |
503 | * "ospke", which is exactly what we want. It shows | |
504 | * that the CPU has PKU, but the OS has not enabled it. | |
505 | * This happens to be exactly how a system would look | |
506 | * if we disabled the config option. | |
507 | */ | |
508 | pr_info("x86: 'nopku' specified, disabling Memory Protection Keys\n"); | |
509 | pku_disabled = true; | |
510 | return 1; | |
511 | } | |
512 | __setup("nopku", setup_disable_pku); | |
513 | #endif /* CONFIG_X86_64 */ | |
514 | ||
515 | /* | |
516 | * Some CPU features depend on higher CPUID levels, which may not always | |
517 | * be available due to CPUID level capping or broken virtualization | |
518 | * software. Add those features to this table to auto-disable them. | |
519 | */ | |
520 | struct cpuid_dependent_feature { | |
521 | u32 feature; | |
522 | u32 level; | |
523 | }; | |
524 | ||
525 | static const struct cpuid_dependent_feature | |
526 | cpuid_dependent_features[] = { | |
527 | { X86_FEATURE_MWAIT, 0x00000005 }, | |
528 | { X86_FEATURE_DCA, 0x00000009 }, | |
529 | { X86_FEATURE_XSAVE, 0x0000000d }, | |
530 | { 0, 0 } | |
531 | }; | |
532 | ||
533 | static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn) | |
534 | { | |
535 | const struct cpuid_dependent_feature *df; | |
536 | ||
537 | for (df = cpuid_dependent_features; df->feature; df++) { | |
538 | ||
539 | if (!cpu_has(c, df->feature)) | |
540 | continue; | |
541 | /* | |
542 | * Note: cpuid_level is set to -1 if unavailable, but | |
543 | * extended_extended_level is set to 0 if unavailable | |
544 | * and the legitimate extended levels are all negative | |
545 | * when signed; hence the weird messing around with | |
546 | * signs here... | |
547 | */ | |
548 | if (!((s32)df->level < 0 ? | |
549 | (u32)df->level > (u32)c->extended_cpuid_level : | |
550 | (s32)df->level > (s32)c->cpuid_level)) | |
551 | continue; | |
552 | ||
553 | clear_cpu_cap(c, df->feature); | |
554 | if (!warn) | |
555 | continue; | |
556 | ||
557 | pr_warn("CPU: CPU feature " X86_CAP_FMT " disabled, no CPUID level 0x%x\n", | |
558 | x86_cap_flag(df->feature), df->level); | |
559 | } | |
560 | } | |
561 | ||
562 | /* | |
563 | * Naming convention should be: <Name> [(<Codename>)] | |
564 | * This table only is used unless init_<vendor>() below doesn't set it; | |
565 | * in particular, if CPUID levels 0x80000002..4 are supported, this | |
566 | * isn't used | |
567 | */ | |
568 | ||
569 | /* Look up CPU names by table lookup. */ | |
570 | static const char *table_lookup_model(struct cpuinfo_x86 *c) | |
571 | { | |
572 | #ifdef CONFIG_X86_32 | |
573 | const struct legacy_cpu_model_info *info; | |
574 | ||
575 | if (c->x86_model >= 16) | |
576 | return NULL; /* Range check */ | |
577 | ||
578 | if (!this_cpu) | |
579 | return NULL; | |
580 | ||
581 | info = this_cpu->legacy_models; | |
582 | ||
583 | while (info->family) { | |
584 | if (info->family == c->x86) | |
585 | return info->model_names[c->x86_model]; | |
586 | info++; | |
587 | } | |
588 | #endif | |
589 | return NULL; /* Not found */ | |
590 | } | |
591 | ||
592 | /* Aligned to unsigned long to avoid split lock in atomic bitmap ops */ | |
593 | __u32 cpu_caps_cleared[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long)); | |
594 | __u32 cpu_caps_set[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long)); | |
595 | ||
596 | void load_percpu_segment(int cpu) | |
597 | { | |
598 | #ifdef CONFIG_X86_32 | |
599 | loadsegment(fs, __KERNEL_PERCPU); | |
600 | #else | |
601 | __loadsegment_simple(gs, 0); | |
602 | wrmsrl(MSR_GS_BASE, cpu_kernelmode_gs_base(cpu)); | |
603 | #endif | |
604 | } | |
605 | ||
606 | #ifdef CONFIG_X86_32 | |
607 | /* The 32-bit entry code needs to find cpu_entry_area. */ | |
608 | DEFINE_PER_CPU(struct cpu_entry_area *, cpu_entry_area); | |
609 | #endif | |
610 | ||
611 | /* Load the original GDT from the per-cpu structure */ | |
612 | void load_direct_gdt(int cpu) | |
613 | { | |
614 | struct desc_ptr gdt_descr; | |
615 | ||
616 | gdt_descr.address = (long)get_cpu_gdt_rw(cpu); | |
617 | gdt_descr.size = GDT_SIZE - 1; | |
618 | load_gdt(&gdt_descr); | |
619 | } | |
620 | EXPORT_SYMBOL_GPL(load_direct_gdt); | |
621 | ||
622 | /* Load a fixmap remapping of the per-cpu GDT */ | |
623 | void load_fixmap_gdt(int cpu) | |
624 | { | |
625 | struct desc_ptr gdt_descr; | |
626 | ||
627 | gdt_descr.address = (long)get_cpu_gdt_ro(cpu); | |
628 | gdt_descr.size = GDT_SIZE - 1; | |
629 | load_gdt(&gdt_descr); | |
630 | } | |
631 | EXPORT_SYMBOL_GPL(load_fixmap_gdt); | |
632 | ||
633 | /* | |
634 | * Current gdt points %fs at the "master" per-cpu area: after this, | |
635 | * it's on the real one. | |
636 | */ | |
637 | void switch_to_new_gdt(int cpu) | |
638 | { | |
639 | /* Load the original GDT */ | |
640 | load_direct_gdt(cpu); | |
641 | /* Reload the per-cpu base */ | |
642 | load_percpu_segment(cpu); | |
643 | } | |
644 | ||
645 | static const struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {}; | |
646 | ||
647 | static void get_model_name(struct cpuinfo_x86 *c) | |
648 | { | |
649 | unsigned int *v; | |
650 | char *p, *q, *s; | |
651 | ||
652 | if (c->extended_cpuid_level < 0x80000004) | |
653 | return; | |
654 | ||
655 | v = (unsigned int *)c->x86_model_id; | |
656 | cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]); | |
657 | cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]); | |
658 | cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]); | |
659 | c->x86_model_id[48] = 0; | |
660 | ||
661 | /* Trim whitespace */ | |
662 | p = q = s = &c->x86_model_id[0]; | |
663 | ||
664 | while (*p == ' ') | |
665 | p++; | |
666 | ||
667 | while (*p) { | |
668 | /* Note the last non-whitespace index */ | |
669 | if (!isspace(*p)) | |
670 | s = q; | |
671 | ||
672 | *q++ = *p++; | |
673 | } | |
674 | ||
675 | *(s + 1) = '\0'; | |
676 | } | |
677 | ||
678 | void detect_num_cpu_cores(struct cpuinfo_x86 *c) | |
679 | { | |
680 | unsigned int eax, ebx, ecx, edx; | |
681 | ||
682 | c->x86_max_cores = 1; | |
683 | if (!IS_ENABLED(CONFIG_SMP) || c->cpuid_level < 4) | |
684 | return; | |
685 | ||
686 | cpuid_count(4, 0, &eax, &ebx, &ecx, &edx); | |
687 | if (eax & 0x1f) | |
688 | c->x86_max_cores = (eax >> 26) + 1; | |
689 | } | |
690 | ||
691 | void cpu_detect_cache_sizes(struct cpuinfo_x86 *c) | |
692 | { | |
693 | unsigned int n, dummy, ebx, ecx, edx, l2size; | |
694 | ||
695 | n = c->extended_cpuid_level; | |
696 | ||
697 | if (n >= 0x80000005) { | |
698 | cpuid(0x80000005, &dummy, &ebx, &ecx, &edx); | |
699 | c->x86_cache_size = (ecx>>24) + (edx>>24); | |
700 | #ifdef CONFIG_X86_64 | |
701 | /* On K8 L1 TLB is inclusive, so don't count it */ | |
702 | c->x86_tlbsize = 0; | |
703 | #endif | |
704 | } | |
705 | ||
706 | if (n < 0x80000006) /* Some chips just has a large L1. */ | |
707 | return; | |
708 | ||
709 | cpuid(0x80000006, &dummy, &ebx, &ecx, &edx); | |
710 | l2size = ecx >> 16; | |
711 | ||
712 | #ifdef CONFIG_X86_64 | |
713 | c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff); | |
714 | #else | |
715 | /* do processor-specific cache resizing */ | |
716 | if (this_cpu->legacy_cache_size) | |
717 | l2size = this_cpu->legacy_cache_size(c, l2size); | |
718 | ||
719 | /* Allow user to override all this if necessary. */ | |
720 | if (cachesize_override != -1) | |
721 | l2size = cachesize_override; | |
722 | ||
723 | if (l2size == 0) | |
724 | return; /* Again, no L2 cache is possible */ | |
725 | #endif | |
726 | ||
727 | c->x86_cache_size = l2size; | |
728 | } | |
729 | ||
730 | u16 __read_mostly tlb_lli_4k[NR_INFO]; | |
731 | u16 __read_mostly tlb_lli_2m[NR_INFO]; | |
732 | u16 __read_mostly tlb_lli_4m[NR_INFO]; | |
733 | u16 __read_mostly tlb_lld_4k[NR_INFO]; | |
734 | u16 __read_mostly tlb_lld_2m[NR_INFO]; | |
735 | u16 __read_mostly tlb_lld_4m[NR_INFO]; | |
736 | u16 __read_mostly tlb_lld_1g[NR_INFO]; | |
737 | ||
738 | static void cpu_detect_tlb(struct cpuinfo_x86 *c) | |
739 | { | |
740 | if (this_cpu->c_detect_tlb) | |
741 | this_cpu->c_detect_tlb(c); | |
742 | ||
743 | pr_info("Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n", | |
744 | tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES], | |
745 | tlb_lli_4m[ENTRIES]); | |
746 | ||
747 | pr_info("Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n", | |
748 | tlb_lld_4k[ENTRIES], tlb_lld_2m[ENTRIES], | |
749 | tlb_lld_4m[ENTRIES], tlb_lld_1g[ENTRIES]); | |
750 | } | |
751 | ||
752 | int detect_ht_early(struct cpuinfo_x86 *c) | |
753 | { | |
754 | #ifdef CONFIG_SMP | |
755 | u32 eax, ebx, ecx, edx; | |
756 | ||
757 | if (!cpu_has(c, X86_FEATURE_HT)) | |
758 | return -1; | |
759 | ||
760 | if (cpu_has(c, X86_FEATURE_CMP_LEGACY)) | |
761 | return -1; | |
762 | ||
763 | if (cpu_has(c, X86_FEATURE_XTOPOLOGY)) | |
764 | return -1; | |
765 | ||
766 | cpuid(1, &eax, &ebx, &ecx, &edx); | |
767 | ||
768 | smp_num_siblings = (ebx & 0xff0000) >> 16; | |
769 | if (smp_num_siblings == 1) | |
770 | pr_info_once("CPU0: Hyper-Threading is disabled\n"); | |
771 | #endif | |
772 | return 0; | |
773 | } | |
774 | ||
775 | void detect_ht(struct cpuinfo_x86 *c) | |
776 | { | |
777 | #ifdef CONFIG_SMP | |
778 | int index_msb, core_bits; | |
779 | ||
780 | if (detect_ht_early(c) < 0) | |
781 | return; | |
782 | ||
783 | index_msb = get_count_order(smp_num_siblings); | |
784 | c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb); | |
785 | ||
786 | smp_num_siblings = smp_num_siblings / c->x86_max_cores; | |
787 | ||
788 | index_msb = get_count_order(smp_num_siblings); | |
789 | ||
790 | core_bits = get_count_order(c->x86_max_cores); | |
791 | ||
792 | c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) & | |
793 | ((1 << core_bits) - 1); | |
794 | #endif | |
795 | } | |
796 | ||
797 | static void get_cpu_vendor(struct cpuinfo_x86 *c) | |
798 | { | |
799 | char *v = c->x86_vendor_id; | |
800 | int i; | |
801 | ||
802 | for (i = 0; i < X86_VENDOR_NUM; i++) { | |
803 | if (!cpu_devs[i]) | |
804 | break; | |
805 | ||
806 | if (!strcmp(v, cpu_devs[i]->c_ident[0]) || | |
807 | (cpu_devs[i]->c_ident[1] && | |
808 | !strcmp(v, cpu_devs[i]->c_ident[1]))) { | |
809 | ||
810 | this_cpu = cpu_devs[i]; | |
811 | c->x86_vendor = this_cpu->c_x86_vendor; | |
812 | return; | |
813 | } | |
814 | } | |
815 | ||
816 | pr_err_once("CPU: vendor_id '%s' unknown, using generic init.\n" \ | |
817 | "CPU: Your system may be unstable.\n", v); | |
818 | ||
819 | c->x86_vendor = X86_VENDOR_UNKNOWN; | |
820 | this_cpu = &default_cpu; | |
821 | } | |
822 | ||
823 | void cpu_detect(struct cpuinfo_x86 *c) | |
824 | { | |
825 | /* Get vendor name */ | |
826 | cpuid(0x00000000, (unsigned int *)&c->cpuid_level, | |
827 | (unsigned int *)&c->x86_vendor_id[0], | |
828 | (unsigned int *)&c->x86_vendor_id[8], | |
829 | (unsigned int *)&c->x86_vendor_id[4]); | |
830 | ||
831 | c->x86 = 4; | |
832 | /* Intel-defined flags: level 0x00000001 */ | |
833 | if (c->cpuid_level >= 0x00000001) { | |
834 | u32 junk, tfms, cap0, misc; | |
835 | ||
836 | cpuid(0x00000001, &tfms, &misc, &junk, &cap0); | |
837 | c->x86 = x86_family(tfms); | |
838 | c->x86_model = x86_model(tfms); | |
839 | c->x86_stepping = x86_stepping(tfms); | |
840 | ||
841 | if (cap0 & (1<<19)) { | |
842 | c->x86_clflush_size = ((misc >> 8) & 0xff) * 8; | |
843 | c->x86_cache_alignment = c->x86_clflush_size; | |
844 | } | |
845 | } | |
846 | } | |
847 | ||
848 | static void apply_forced_caps(struct cpuinfo_x86 *c) | |
849 | { | |
850 | int i; | |
851 | ||
852 | for (i = 0; i < NCAPINTS + NBUGINTS; i++) { | |
853 | c->x86_capability[i] &= ~cpu_caps_cleared[i]; | |
854 | c->x86_capability[i] |= cpu_caps_set[i]; | |
855 | } | |
856 | } | |
857 | ||
858 | static void init_speculation_control(struct cpuinfo_x86 *c) | |
859 | { | |
860 | /* | |
861 | * The Intel SPEC_CTRL CPUID bit implies IBRS and IBPB support, | |
862 | * and they also have a different bit for STIBP support. Also, | |
863 | * a hypervisor might have set the individual AMD bits even on | |
864 | * Intel CPUs, for finer-grained selection of what's available. | |
865 | */ | |
866 | if (cpu_has(c, X86_FEATURE_SPEC_CTRL)) { | |
867 | set_cpu_cap(c, X86_FEATURE_IBRS); | |
868 | set_cpu_cap(c, X86_FEATURE_IBPB); | |
869 | set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL); | |
870 | } | |
871 | ||
872 | if (cpu_has(c, X86_FEATURE_INTEL_STIBP)) | |
873 | set_cpu_cap(c, X86_FEATURE_STIBP); | |
874 | ||
875 | if (cpu_has(c, X86_FEATURE_SPEC_CTRL_SSBD) || | |
876 | cpu_has(c, X86_FEATURE_VIRT_SSBD)) | |
877 | set_cpu_cap(c, X86_FEATURE_SSBD); | |
878 | ||
879 | if (cpu_has(c, X86_FEATURE_AMD_IBRS)) { | |
880 | set_cpu_cap(c, X86_FEATURE_IBRS); | |
881 | set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL); | |
882 | } | |
883 | ||
884 | if (cpu_has(c, X86_FEATURE_AMD_IBPB)) | |
885 | set_cpu_cap(c, X86_FEATURE_IBPB); | |
886 | ||
887 | if (cpu_has(c, X86_FEATURE_AMD_STIBP)) { | |
888 | set_cpu_cap(c, X86_FEATURE_STIBP); | |
889 | set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL); | |
890 | } | |
891 | ||
892 | if (cpu_has(c, X86_FEATURE_AMD_SSBD)) { | |
893 | set_cpu_cap(c, X86_FEATURE_SSBD); | |
894 | set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL); | |
895 | clear_cpu_cap(c, X86_FEATURE_VIRT_SSBD); | |
896 | } | |
897 | } | |
898 | ||
899 | void get_cpu_cap(struct cpuinfo_x86 *c) | |
900 | { | |
901 | u32 eax, ebx, ecx, edx; | |
902 | ||
903 | /* Intel-defined flags: level 0x00000001 */ | |
904 | if (c->cpuid_level >= 0x00000001) { | |
905 | cpuid(0x00000001, &eax, &ebx, &ecx, &edx); | |
906 | ||
907 | c->x86_capability[CPUID_1_ECX] = ecx; | |
908 | c->x86_capability[CPUID_1_EDX] = edx; | |
909 | } | |
910 | ||
911 | /* Thermal and Power Management Leaf: level 0x00000006 (eax) */ | |
912 | if (c->cpuid_level >= 0x00000006) | |
913 | c->x86_capability[CPUID_6_EAX] = cpuid_eax(0x00000006); | |
914 | ||
915 | /* Additional Intel-defined flags: level 0x00000007 */ | |
916 | if (c->cpuid_level >= 0x00000007) { | |
917 | cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx); | |
918 | c->x86_capability[CPUID_7_0_EBX] = ebx; | |
919 | c->x86_capability[CPUID_7_ECX] = ecx; | |
920 | c->x86_capability[CPUID_7_EDX] = edx; | |
921 | ||
922 | /* Check valid sub-leaf index before accessing it */ | |
923 | if (eax >= 1) { | |
924 | cpuid_count(0x00000007, 1, &eax, &ebx, &ecx, &edx); | |
925 | c->x86_capability[CPUID_7_1_EAX] = eax; | |
926 | } | |
927 | } | |
928 | ||
929 | /* Extended state features: level 0x0000000d */ | |
930 | if (c->cpuid_level >= 0x0000000d) { | |
931 | cpuid_count(0x0000000d, 1, &eax, &ebx, &ecx, &edx); | |
932 | ||
933 | c->x86_capability[CPUID_D_1_EAX] = eax; | |
934 | } | |
935 | ||
936 | /* AMD-defined flags: level 0x80000001 */ | |
937 | eax = cpuid_eax(0x80000000); | |
938 | c->extended_cpuid_level = eax; | |
939 | ||
940 | if ((eax & 0xffff0000) == 0x80000000) { | |
941 | if (eax >= 0x80000001) { | |
942 | cpuid(0x80000001, &eax, &ebx, &ecx, &edx); | |
943 | ||
944 | c->x86_capability[CPUID_8000_0001_ECX] = ecx; | |
945 | c->x86_capability[CPUID_8000_0001_EDX] = edx; | |
946 | } | |
947 | } | |
948 | ||
949 | if (c->extended_cpuid_level >= 0x80000007) { | |
950 | cpuid(0x80000007, &eax, &ebx, &ecx, &edx); | |
951 | ||
952 | c->x86_capability[CPUID_8000_0007_EBX] = ebx; | |
953 | c->x86_power = edx; | |
954 | } | |
955 | ||
956 | if (c->extended_cpuid_level >= 0x80000008) { | |
957 | cpuid(0x80000008, &eax, &ebx, &ecx, &edx); | |
958 | c->x86_capability[CPUID_8000_0008_EBX] = ebx; | |
959 | } | |
960 | ||
961 | if (c->extended_cpuid_level >= 0x8000000a) | |
962 | c->x86_capability[CPUID_8000_000A_EDX] = cpuid_edx(0x8000000a); | |
963 | ||
964 | if (c->extended_cpuid_level >= 0x8000001f) | |
965 | c->x86_capability[CPUID_8000_001F_EAX] = cpuid_eax(0x8000001f); | |
966 | ||
967 | init_scattered_cpuid_features(c); | |
968 | init_speculation_control(c); | |
969 | ||
970 | /* | |
971 | * Clear/Set all flags overridden by options, after probe. | |
972 | * This needs to happen each time we re-probe, which may happen | |
973 | * several times during CPU initialization. | |
974 | */ | |
975 | apply_forced_caps(c); | |
976 | } | |
977 | ||
978 | void get_cpu_address_sizes(struct cpuinfo_x86 *c) | |
979 | { | |
980 | u32 eax, ebx, ecx, edx; | |
981 | ||
982 | if (c->extended_cpuid_level >= 0x80000008) { | |
983 | cpuid(0x80000008, &eax, &ebx, &ecx, &edx); | |
984 | ||
985 | c->x86_virt_bits = (eax >> 8) & 0xff; | |
986 | c->x86_phys_bits = eax & 0xff; | |
987 | } | |
988 | #ifdef CONFIG_X86_32 | |
989 | else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36)) | |
990 | c->x86_phys_bits = 36; | |
991 | #endif | |
992 | c->x86_cache_bits = c->x86_phys_bits; | |
993 | } | |
994 | ||
995 | static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c) | |
996 | { | |
997 | #ifdef CONFIG_X86_32 | |
998 | int i; | |
999 | ||
1000 | /* | |
1001 | * First of all, decide if this is a 486 or higher | |
1002 | * It's a 486 if we can modify the AC flag | |
1003 | */ | |
1004 | if (flag_is_changeable_p(X86_EFLAGS_AC)) | |
1005 | c->x86 = 4; | |
1006 | else | |
1007 | c->x86 = 3; | |
1008 | ||
1009 | for (i = 0; i < X86_VENDOR_NUM; i++) | |
1010 | if (cpu_devs[i] && cpu_devs[i]->c_identify) { | |
1011 | c->x86_vendor_id[0] = 0; | |
1012 | cpu_devs[i]->c_identify(c); | |
1013 | if (c->x86_vendor_id[0]) { | |
1014 | get_cpu_vendor(c); | |
1015 | break; | |
1016 | } | |
1017 | } | |
1018 | #endif | |
1019 | } | |
1020 | ||
1021 | #define NO_SPECULATION BIT(0) | |
1022 | #define NO_MELTDOWN BIT(1) | |
1023 | #define NO_SSB BIT(2) | |
1024 | #define NO_L1TF BIT(3) | |
1025 | #define NO_MDS BIT(4) | |
1026 | #define MSBDS_ONLY BIT(5) | |
1027 | #define NO_SWAPGS BIT(6) | |
1028 | #define NO_ITLB_MULTIHIT BIT(7) | |
1029 | #define NO_SPECTRE_V2 BIT(8) | |
1030 | ||
1031 | #define VULNWL(vendor, family, model, whitelist) \ | |
1032 | X86_MATCH_VENDOR_FAM_MODEL(vendor, family, model, whitelist) | |
1033 | ||
1034 | #define VULNWL_INTEL(model, whitelist) \ | |
1035 | VULNWL(INTEL, 6, INTEL_FAM6_##model, whitelist) | |
1036 | ||
1037 | #define VULNWL_AMD(family, whitelist) \ | |
1038 | VULNWL(AMD, family, X86_MODEL_ANY, whitelist) | |
1039 | ||
1040 | #define VULNWL_HYGON(family, whitelist) \ | |
1041 | VULNWL(HYGON, family, X86_MODEL_ANY, whitelist) | |
1042 | ||
1043 | static const __initconst struct x86_cpu_id cpu_vuln_whitelist[] = { | |
1044 | VULNWL(ANY, 4, X86_MODEL_ANY, NO_SPECULATION), | |
1045 | VULNWL(CENTAUR, 5, X86_MODEL_ANY, NO_SPECULATION), | |
1046 | VULNWL(INTEL, 5, X86_MODEL_ANY, NO_SPECULATION), | |
1047 | VULNWL(NSC, 5, X86_MODEL_ANY, NO_SPECULATION), | |
1048 | ||
1049 | /* Intel Family 6 */ | |
1050 | VULNWL_INTEL(ATOM_SALTWELL, NO_SPECULATION | NO_ITLB_MULTIHIT), | |
1051 | VULNWL_INTEL(ATOM_SALTWELL_TABLET, NO_SPECULATION | NO_ITLB_MULTIHIT), | |
1052 | VULNWL_INTEL(ATOM_SALTWELL_MID, NO_SPECULATION | NO_ITLB_MULTIHIT), | |
1053 | VULNWL_INTEL(ATOM_BONNELL, NO_SPECULATION | NO_ITLB_MULTIHIT), | |
1054 | VULNWL_INTEL(ATOM_BONNELL_MID, NO_SPECULATION | NO_ITLB_MULTIHIT), | |
1055 | ||
1056 | VULNWL_INTEL(ATOM_SILVERMONT, NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT), | |
1057 | VULNWL_INTEL(ATOM_SILVERMONT_D, NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT), | |
1058 | VULNWL_INTEL(ATOM_SILVERMONT_MID, NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT), | |
1059 | VULNWL_INTEL(ATOM_AIRMONT, NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT), | |
1060 | VULNWL_INTEL(XEON_PHI_KNL, NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT), | |
1061 | VULNWL_INTEL(XEON_PHI_KNM, NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT), | |
1062 | ||
1063 | VULNWL_INTEL(CORE_YONAH, NO_SSB), | |
1064 | ||
1065 | VULNWL_INTEL(ATOM_AIRMONT_MID, NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT), | |
1066 | VULNWL_INTEL(ATOM_AIRMONT_NP, NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT), | |
1067 | ||
1068 | VULNWL_INTEL(ATOM_GOLDMONT, NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT), | |
1069 | VULNWL_INTEL(ATOM_GOLDMONT_D, NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT), | |
1070 | VULNWL_INTEL(ATOM_GOLDMONT_PLUS, NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT), | |
1071 | ||
1072 | /* | |
1073 | * Technically, swapgs isn't serializing on AMD (despite it previously | |
1074 | * being documented as such in the APM). But according to AMD, %gs is | |
1075 | * updated non-speculatively, and the issuing of %gs-relative memory | |
1076 | * operands will be blocked until the %gs update completes, which is | |
1077 | * good enough for our purposes. | |
1078 | */ | |
1079 | ||
1080 | VULNWL_INTEL(ATOM_TREMONT_D, NO_ITLB_MULTIHIT), | |
1081 | ||
1082 | /* AMD Family 0xf - 0x12 */ | |
1083 | VULNWL_AMD(0x0f, NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT), | |
1084 | VULNWL_AMD(0x10, NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT), | |
1085 | VULNWL_AMD(0x11, NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT), | |
1086 | VULNWL_AMD(0x12, NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT), | |
1087 | ||
1088 | /* FAMILY_ANY must be last, otherwise 0x0f - 0x12 matches won't work */ | |
1089 | VULNWL_AMD(X86_FAMILY_ANY, NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT), | |
1090 | VULNWL_HYGON(X86_FAMILY_ANY, NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT), | |
1091 | ||
1092 | /* Zhaoxin Family 7 */ | |
1093 | VULNWL(CENTAUR, 7, X86_MODEL_ANY, NO_SPECTRE_V2 | NO_SWAPGS), | |
1094 | VULNWL(ZHAOXIN, 7, X86_MODEL_ANY, NO_SPECTRE_V2 | NO_SWAPGS), | |
1095 | {} | |
1096 | }; | |
1097 | ||
1098 | #define VULNBL_INTEL_STEPPINGS(model, steppings, issues) \ | |
1099 | X86_MATCH_VENDOR_FAM_MODEL_STEPPINGS_FEATURE(INTEL, 6, \ | |
1100 | INTEL_FAM6_##model, steppings, \ | |
1101 | X86_FEATURE_ANY, issues) | |
1102 | ||
1103 | #define SRBDS BIT(0) | |
1104 | /* CPU is affected by X86_BUG_MMIO_STALE_DATA */ | |
1105 | #define MMIO BIT(1) | |
1106 | ||
1107 | static const struct x86_cpu_id cpu_vuln_blacklist[] __initconst = { | |
1108 | VULNBL_INTEL_STEPPINGS(IVYBRIDGE, X86_STEPPING_ANY, SRBDS), | |
1109 | VULNBL_INTEL_STEPPINGS(HASWELL, X86_STEPPING_ANY, SRBDS), | |
1110 | VULNBL_INTEL_STEPPINGS(HASWELL_L, X86_STEPPING_ANY, SRBDS), | |
1111 | VULNBL_INTEL_STEPPINGS(HASWELL_G, X86_STEPPING_ANY, SRBDS), | |
1112 | VULNBL_INTEL_STEPPINGS(HASWELL_X, BIT(2) | BIT(4), MMIO), | |
1113 | VULNBL_INTEL_STEPPINGS(BROADWELL_D, X86_STEPPINGS(0x3, 0x5), MMIO), | |
1114 | VULNBL_INTEL_STEPPINGS(BROADWELL_G, X86_STEPPING_ANY, SRBDS), | |
1115 | VULNBL_INTEL_STEPPINGS(BROADWELL_X, X86_STEPPING_ANY, MMIO), | |
1116 | VULNBL_INTEL_STEPPINGS(BROADWELL, X86_STEPPING_ANY, SRBDS), | |
1117 | VULNBL_INTEL_STEPPINGS(SKYLAKE_L, X86_STEPPINGS(0x3, 0x3), SRBDS | MMIO), | |
1118 | VULNBL_INTEL_STEPPINGS(SKYLAKE_L, X86_STEPPING_ANY, SRBDS), | |
1119 | VULNBL_INTEL_STEPPINGS(SKYLAKE_X, BIT(3) | BIT(4) | BIT(6) | | |
1120 | BIT(7) | BIT(0xB), MMIO), | |
1121 | VULNBL_INTEL_STEPPINGS(SKYLAKE, X86_STEPPINGS(0x3, 0x3), SRBDS | MMIO), | |
1122 | VULNBL_INTEL_STEPPINGS(SKYLAKE, X86_STEPPING_ANY, SRBDS), | |
1123 | VULNBL_INTEL_STEPPINGS(KABYLAKE_L, X86_STEPPINGS(0x9, 0xC), SRBDS | MMIO), | |
1124 | VULNBL_INTEL_STEPPINGS(KABYLAKE_L, X86_STEPPINGS(0x0, 0x8), SRBDS), | |
1125 | VULNBL_INTEL_STEPPINGS(KABYLAKE, X86_STEPPINGS(0x9, 0xD), SRBDS | MMIO), | |
1126 | VULNBL_INTEL_STEPPINGS(KABYLAKE, X86_STEPPINGS(0x0, 0x8), SRBDS), | |
1127 | VULNBL_INTEL_STEPPINGS(ICELAKE_L, X86_STEPPINGS(0x5, 0x5), MMIO), | |
1128 | VULNBL_INTEL_STEPPINGS(ICELAKE_D, X86_STEPPINGS(0x1, 0x1), MMIO), | |
1129 | VULNBL_INTEL_STEPPINGS(ICELAKE_X, X86_STEPPINGS(0x4, 0x6), MMIO), | |
1130 | VULNBL_INTEL_STEPPINGS(COMETLAKE, BIT(2) | BIT(3) | BIT(5), MMIO), | |
1131 | VULNBL_INTEL_STEPPINGS(COMETLAKE_L, X86_STEPPINGS(0x0, 0x1), MMIO), | |
1132 | VULNBL_INTEL_STEPPINGS(LAKEFIELD, X86_STEPPINGS(0x1, 0x1), MMIO), | |
1133 | VULNBL_INTEL_STEPPINGS(ROCKETLAKE, X86_STEPPINGS(0x1, 0x1), MMIO), | |
1134 | VULNBL_INTEL_STEPPINGS(ATOM_TREMONT, X86_STEPPINGS(0x1, 0x1), MMIO), | |
1135 | VULNBL_INTEL_STEPPINGS(ATOM_TREMONT_D, X86_STEPPING_ANY, MMIO), | |
1136 | VULNBL_INTEL_STEPPINGS(ATOM_TREMONT_L, X86_STEPPINGS(0x0, 0x0), MMIO), | |
1137 | {} | |
1138 | }; | |
1139 | ||
1140 | static bool __init cpu_matches(const struct x86_cpu_id *table, unsigned long which) | |
1141 | { | |
1142 | const struct x86_cpu_id *m = x86_match_cpu(table); | |
1143 | ||
1144 | return m && !!(m->driver_data & which); | |
1145 | } | |
1146 | ||
1147 | u64 x86_read_arch_cap_msr(void) | |
1148 | { | |
1149 | u64 ia32_cap = 0; | |
1150 | ||
1151 | if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES)) | |
1152 | rdmsrl(MSR_IA32_ARCH_CAPABILITIES, ia32_cap); | |
1153 | ||
1154 | return ia32_cap; | |
1155 | } | |
1156 | ||
1157 | static bool arch_cap_mmio_immune(u64 ia32_cap) | |
1158 | { | |
1159 | return (ia32_cap & ARCH_CAP_FBSDP_NO && | |
1160 | ia32_cap & ARCH_CAP_PSDP_NO && | |
1161 | ia32_cap & ARCH_CAP_SBDR_SSDP_NO); | |
1162 | } | |
1163 | ||
1164 | static void __init cpu_set_bug_bits(struct cpuinfo_x86 *c) | |
1165 | { | |
1166 | u64 ia32_cap = x86_read_arch_cap_msr(); | |
1167 | ||
1168 | /* Set ITLB_MULTIHIT bug if cpu is not in the whitelist and not mitigated */ | |
1169 | if (!cpu_matches(cpu_vuln_whitelist, NO_ITLB_MULTIHIT) && | |
1170 | !(ia32_cap & ARCH_CAP_PSCHANGE_MC_NO)) | |
1171 | setup_force_cpu_bug(X86_BUG_ITLB_MULTIHIT); | |
1172 | ||
1173 | if (cpu_matches(cpu_vuln_whitelist, NO_SPECULATION)) | |
1174 | return; | |
1175 | ||
1176 | setup_force_cpu_bug(X86_BUG_SPECTRE_V1); | |
1177 | ||
1178 | if (!cpu_matches(cpu_vuln_whitelist, NO_SPECTRE_V2)) | |
1179 | setup_force_cpu_bug(X86_BUG_SPECTRE_V2); | |
1180 | ||
1181 | if (!cpu_matches(cpu_vuln_whitelist, NO_SSB) && | |
1182 | !(ia32_cap & ARCH_CAP_SSB_NO) && | |
1183 | !cpu_has(c, X86_FEATURE_AMD_SSB_NO)) | |
1184 | setup_force_cpu_bug(X86_BUG_SPEC_STORE_BYPASS); | |
1185 | ||
1186 | if (ia32_cap & ARCH_CAP_IBRS_ALL) | |
1187 | setup_force_cpu_cap(X86_FEATURE_IBRS_ENHANCED); | |
1188 | ||
1189 | if (!cpu_matches(cpu_vuln_whitelist, NO_MDS) && | |
1190 | !(ia32_cap & ARCH_CAP_MDS_NO)) { | |
1191 | setup_force_cpu_bug(X86_BUG_MDS); | |
1192 | if (cpu_matches(cpu_vuln_whitelist, MSBDS_ONLY)) | |
1193 | setup_force_cpu_bug(X86_BUG_MSBDS_ONLY); | |
1194 | } | |
1195 | ||
1196 | if (!cpu_matches(cpu_vuln_whitelist, NO_SWAPGS)) | |
1197 | setup_force_cpu_bug(X86_BUG_SWAPGS); | |
1198 | ||
1199 | /* | |
1200 | * When the CPU is not mitigated for TAA (TAA_NO=0) set TAA bug when: | |
1201 | * - TSX is supported or | |
1202 | * - TSX_CTRL is present | |
1203 | * | |
1204 | * TSX_CTRL check is needed for cases when TSX could be disabled before | |
1205 | * the kernel boot e.g. kexec. | |
1206 | * TSX_CTRL check alone is not sufficient for cases when the microcode | |
1207 | * update is not present or running as guest that don't get TSX_CTRL. | |
1208 | */ | |
1209 | if (!(ia32_cap & ARCH_CAP_TAA_NO) && | |
1210 | (cpu_has(c, X86_FEATURE_RTM) || | |
1211 | (ia32_cap & ARCH_CAP_TSX_CTRL_MSR))) | |
1212 | setup_force_cpu_bug(X86_BUG_TAA); | |
1213 | ||
1214 | /* | |
1215 | * SRBDS affects CPUs which support RDRAND or RDSEED and are listed | |
1216 | * in the vulnerability blacklist. | |
1217 | */ | |
1218 | if ((cpu_has(c, X86_FEATURE_RDRAND) || | |
1219 | cpu_has(c, X86_FEATURE_RDSEED)) && | |
1220 | cpu_matches(cpu_vuln_blacklist, SRBDS)) | |
1221 | setup_force_cpu_bug(X86_BUG_SRBDS); | |
1222 | ||
1223 | /* | |
1224 | * Processor MMIO Stale Data bug enumeration | |
1225 | * | |
1226 | * Affected CPU list is generally enough to enumerate the vulnerability, | |
1227 | * but for virtualization case check for ARCH_CAP MSR bits also, VMM may | |
1228 | * not want the guest to enumerate the bug. | |
1229 | */ | |
1230 | if (cpu_matches(cpu_vuln_blacklist, MMIO) && | |
1231 | !arch_cap_mmio_immune(ia32_cap)) | |
1232 | setup_force_cpu_bug(X86_BUG_MMIO_STALE_DATA); | |
1233 | ||
1234 | if (cpu_matches(cpu_vuln_whitelist, NO_MELTDOWN)) | |
1235 | return; | |
1236 | ||
1237 | /* Rogue Data Cache Load? No! */ | |
1238 | if (ia32_cap & ARCH_CAP_RDCL_NO) | |
1239 | return; | |
1240 | ||
1241 | setup_force_cpu_bug(X86_BUG_CPU_MELTDOWN); | |
1242 | ||
1243 | if (cpu_matches(cpu_vuln_whitelist, NO_L1TF)) | |
1244 | return; | |
1245 | ||
1246 | setup_force_cpu_bug(X86_BUG_L1TF); | |
1247 | } | |
1248 | ||
1249 | /* | |
1250 | * The NOPL instruction is supposed to exist on all CPUs of family >= 6; | |
1251 | * unfortunately, that's not true in practice because of early VIA | |
1252 | * chips and (more importantly) broken virtualizers that are not easy | |
1253 | * to detect. In the latter case it doesn't even *fail* reliably, so | |
1254 | * probing for it doesn't even work. Disable it completely on 32-bit | |
1255 | * unless we can find a reliable way to detect all the broken cases. | |
1256 | * Enable it explicitly on 64-bit for non-constant inputs of cpu_has(). | |
1257 | */ | |
1258 | static void detect_nopl(void) | |
1259 | { | |
1260 | #ifdef CONFIG_X86_32 | |
1261 | setup_clear_cpu_cap(X86_FEATURE_NOPL); | |
1262 | #else | |
1263 | setup_force_cpu_cap(X86_FEATURE_NOPL); | |
1264 | #endif | |
1265 | } | |
1266 | ||
1267 | /* | |
1268 | * We parse cpu parameters early because fpu__init_system() is executed | |
1269 | * before parse_early_param(). | |
1270 | */ | |
1271 | static void __init cpu_parse_early_param(void) | |
1272 | { | |
1273 | char arg[128]; | |
1274 | char *argptr = arg; | |
1275 | int arglen, res, bit; | |
1276 | ||
1277 | #ifdef CONFIG_X86_32 | |
1278 | if (cmdline_find_option_bool(boot_command_line, "no387")) | |
1279 | #ifdef CONFIG_MATH_EMULATION | |
1280 | setup_clear_cpu_cap(X86_FEATURE_FPU); | |
1281 | #else | |
1282 | pr_err("Option 'no387' required CONFIG_MATH_EMULATION enabled.\n"); | |
1283 | #endif | |
1284 | ||
1285 | if (cmdline_find_option_bool(boot_command_line, "nofxsr")) | |
1286 | setup_clear_cpu_cap(X86_FEATURE_FXSR); | |
1287 | #endif | |
1288 | ||
1289 | if (cmdline_find_option_bool(boot_command_line, "noxsave")) | |
1290 | setup_clear_cpu_cap(X86_FEATURE_XSAVE); | |
1291 | ||
1292 | if (cmdline_find_option_bool(boot_command_line, "noxsaveopt")) | |
1293 | setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT); | |
1294 | ||
1295 | if (cmdline_find_option_bool(boot_command_line, "noxsaves")) | |
1296 | setup_clear_cpu_cap(X86_FEATURE_XSAVES); | |
1297 | ||
1298 | arglen = cmdline_find_option(boot_command_line, "clearcpuid", arg, sizeof(arg)); | |
1299 | if (arglen <= 0) | |
1300 | return; | |
1301 | ||
1302 | pr_info("Clearing CPUID bits:"); | |
1303 | do { | |
1304 | res = get_option(&argptr, &bit); | |
1305 | if (res == 0 || res == 3) | |
1306 | break; | |
1307 | ||
1308 | /* If the argument was too long, the last bit may be cut off */ | |
1309 | if (res == 1 && arglen >= sizeof(arg)) | |
1310 | break; | |
1311 | ||
1312 | if (bit >= 0 && bit < NCAPINTS * 32) { | |
1313 | pr_cont(" " X86_CAP_FMT, x86_cap_flag(bit)); | |
1314 | setup_clear_cpu_cap(bit); | |
1315 | } | |
1316 | } while (res == 2); | |
1317 | pr_cont("\n"); | |
1318 | } | |
1319 | ||
1320 | /* | |
1321 | * Do minimum CPU detection early. | |
1322 | * Fields really needed: vendor, cpuid_level, family, model, mask, | |
1323 | * cache alignment. | |
1324 | * The others are not touched to avoid unwanted side effects. | |
1325 | * | |
1326 | * WARNING: this function is only called on the boot CPU. Don't add code | |
1327 | * here that is supposed to run on all CPUs. | |
1328 | */ | |
1329 | static void __init early_identify_cpu(struct cpuinfo_x86 *c) | |
1330 | { | |
1331 | #ifdef CONFIG_X86_64 | |
1332 | c->x86_clflush_size = 64; | |
1333 | c->x86_phys_bits = 36; | |
1334 | c->x86_virt_bits = 48; | |
1335 | #else | |
1336 | c->x86_clflush_size = 32; | |
1337 | c->x86_phys_bits = 32; | |
1338 | c->x86_virt_bits = 32; | |
1339 | #endif | |
1340 | c->x86_cache_alignment = c->x86_clflush_size; | |
1341 | ||
1342 | memset(&c->x86_capability, 0, sizeof(c->x86_capability)); | |
1343 | c->extended_cpuid_level = 0; | |
1344 | ||
1345 | if (!have_cpuid_p()) | |
1346 | identify_cpu_without_cpuid(c); | |
1347 | ||
1348 | /* cyrix could have cpuid enabled via c_identify()*/ | |
1349 | if (have_cpuid_p()) { | |
1350 | cpu_detect(c); | |
1351 | get_cpu_vendor(c); | |
1352 | get_cpu_cap(c); | |
1353 | get_cpu_address_sizes(c); | |
1354 | setup_force_cpu_cap(X86_FEATURE_CPUID); | |
1355 | cpu_parse_early_param(); | |
1356 | ||
1357 | if (this_cpu->c_early_init) | |
1358 | this_cpu->c_early_init(c); | |
1359 | ||
1360 | c->cpu_index = 0; | |
1361 | filter_cpuid_features(c, false); | |
1362 | ||
1363 | if (this_cpu->c_bsp_init) | |
1364 | this_cpu->c_bsp_init(c); | |
1365 | } else { | |
1366 | setup_clear_cpu_cap(X86_FEATURE_CPUID); | |
1367 | } | |
1368 | ||
1369 | setup_force_cpu_cap(X86_FEATURE_ALWAYS); | |
1370 | ||
1371 | cpu_set_bug_bits(c); | |
1372 | ||
1373 | sld_setup(c); | |
1374 | ||
1375 | fpu__init_system(c); | |
1376 | ||
1377 | init_sigframe_size(); | |
1378 | ||
1379 | #ifdef CONFIG_X86_32 | |
1380 | /* | |
1381 | * Regardless of whether PCID is enumerated, the SDM says | |
1382 | * that it can't be enabled in 32-bit mode. | |
1383 | */ | |
1384 | setup_clear_cpu_cap(X86_FEATURE_PCID); | |
1385 | #endif | |
1386 | ||
1387 | /* | |
1388 | * Later in the boot process pgtable_l5_enabled() relies on | |
1389 | * cpu_feature_enabled(X86_FEATURE_LA57). If 5-level paging is not | |
1390 | * enabled by this point we need to clear the feature bit to avoid | |
1391 | * false-positives at the later stage. | |
1392 | * | |
1393 | * pgtable_l5_enabled() can be false here for several reasons: | |
1394 | * - 5-level paging is disabled compile-time; | |
1395 | * - it's 32-bit kernel; | |
1396 | * - machine doesn't support 5-level paging; | |
1397 | * - user specified 'no5lvl' in kernel command line. | |
1398 | */ | |
1399 | if (!pgtable_l5_enabled()) | |
1400 | setup_clear_cpu_cap(X86_FEATURE_LA57); | |
1401 | ||
1402 | detect_nopl(); | |
1403 | } | |
1404 | ||
1405 | void __init early_cpu_init(void) | |
1406 | { | |
1407 | const struct cpu_dev *const *cdev; | |
1408 | int count = 0; | |
1409 | ||
1410 | #ifdef CONFIG_PROCESSOR_SELECT | |
1411 | pr_info("KERNEL supported cpus:\n"); | |
1412 | #endif | |
1413 | ||
1414 | for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) { | |
1415 | const struct cpu_dev *cpudev = *cdev; | |
1416 | ||
1417 | if (count >= X86_VENDOR_NUM) | |
1418 | break; | |
1419 | cpu_devs[count] = cpudev; | |
1420 | count++; | |
1421 | ||
1422 | #ifdef CONFIG_PROCESSOR_SELECT | |
1423 | { | |
1424 | unsigned int j; | |
1425 | ||
1426 | for (j = 0; j < 2; j++) { | |
1427 | if (!cpudev->c_ident[j]) | |
1428 | continue; | |
1429 | pr_info(" %s %s\n", cpudev->c_vendor, | |
1430 | cpudev->c_ident[j]); | |
1431 | } | |
1432 | } | |
1433 | #endif | |
1434 | } | |
1435 | early_identify_cpu(&boot_cpu_data); | |
1436 | } | |
1437 | ||
1438 | static bool detect_null_seg_behavior(void) | |
1439 | { | |
1440 | /* | |
1441 | * Empirically, writing zero to a segment selector on AMD does | |
1442 | * not clear the base, whereas writing zero to a segment | |
1443 | * selector on Intel does clear the base. Intel's behavior | |
1444 | * allows slightly faster context switches in the common case | |
1445 | * where GS is unused by the prev and next threads. | |
1446 | * | |
1447 | * Since neither vendor documents this anywhere that I can see, | |
1448 | * detect it directly instead of hard-coding the choice by | |
1449 | * vendor. | |
1450 | * | |
1451 | * I've designated AMD's behavior as the "bug" because it's | |
1452 | * counterintuitive and less friendly. | |
1453 | */ | |
1454 | ||
1455 | unsigned long old_base, tmp; | |
1456 | rdmsrl(MSR_FS_BASE, old_base); | |
1457 | wrmsrl(MSR_FS_BASE, 1); | |
1458 | loadsegment(fs, 0); | |
1459 | rdmsrl(MSR_FS_BASE, tmp); | |
1460 | wrmsrl(MSR_FS_BASE, old_base); | |
1461 | return tmp == 0; | |
1462 | } | |
1463 | ||
1464 | void check_null_seg_clears_base(struct cpuinfo_x86 *c) | |
1465 | { | |
1466 | /* BUG_NULL_SEG is only relevant with 64bit userspace */ | |
1467 | if (!IS_ENABLED(CONFIG_X86_64)) | |
1468 | return; | |
1469 | ||
1470 | /* Zen3 CPUs advertise Null Selector Clears Base in CPUID. */ | |
1471 | if (c->extended_cpuid_level >= 0x80000021 && | |
1472 | cpuid_eax(0x80000021) & BIT(6)) | |
1473 | return; | |
1474 | ||
1475 | /* | |
1476 | * CPUID bit above wasn't set. If this kernel is still running | |
1477 | * as a HV guest, then the HV has decided not to advertize | |
1478 | * that CPUID bit for whatever reason. For example, one | |
1479 | * member of the migration pool might be vulnerable. Which | |
1480 | * means, the bug is present: set the BUG flag and return. | |
1481 | */ | |
1482 | if (cpu_has(c, X86_FEATURE_HYPERVISOR)) { | |
1483 | set_cpu_bug(c, X86_BUG_NULL_SEG); | |
1484 | return; | |
1485 | } | |
1486 | ||
1487 | /* | |
1488 | * Zen2 CPUs also have this behaviour, but no CPUID bit. | |
1489 | * 0x18 is the respective family for Hygon. | |
1490 | */ | |
1491 | if ((c->x86 == 0x17 || c->x86 == 0x18) && | |
1492 | detect_null_seg_behavior()) | |
1493 | return; | |
1494 | ||
1495 | /* All the remaining ones are affected */ | |
1496 | set_cpu_bug(c, X86_BUG_NULL_SEG); | |
1497 | } | |
1498 | ||
1499 | static void generic_identify(struct cpuinfo_x86 *c) | |
1500 | { | |
1501 | c->extended_cpuid_level = 0; | |
1502 | ||
1503 | if (!have_cpuid_p()) | |
1504 | identify_cpu_without_cpuid(c); | |
1505 | ||
1506 | /* cyrix could have cpuid enabled via c_identify()*/ | |
1507 | if (!have_cpuid_p()) | |
1508 | return; | |
1509 | ||
1510 | cpu_detect(c); | |
1511 | ||
1512 | get_cpu_vendor(c); | |
1513 | ||
1514 | get_cpu_cap(c); | |
1515 | ||
1516 | get_cpu_address_sizes(c); | |
1517 | ||
1518 | if (c->cpuid_level >= 0x00000001) { | |
1519 | c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF; | |
1520 | #ifdef CONFIG_X86_32 | |
1521 | # ifdef CONFIG_SMP | |
1522 | c->apicid = apic->phys_pkg_id(c->initial_apicid, 0); | |
1523 | # else | |
1524 | c->apicid = c->initial_apicid; | |
1525 | # endif | |
1526 | #endif | |
1527 | c->phys_proc_id = c->initial_apicid; | |
1528 | } | |
1529 | ||
1530 | get_model_name(c); /* Default name */ | |
1531 | ||
1532 | /* | |
1533 | * ESPFIX is a strange bug. All real CPUs have it. Paravirt | |
1534 | * systems that run Linux at CPL > 0 may or may not have the | |
1535 | * issue, but, even if they have the issue, there's absolutely | |
1536 | * nothing we can do about it because we can't use the real IRET | |
1537 | * instruction. | |
1538 | * | |
1539 | * NB: For the time being, only 32-bit kernels support | |
1540 | * X86_BUG_ESPFIX as such. 64-bit kernels directly choose | |
1541 | * whether to apply espfix using paravirt hooks. If any | |
1542 | * non-paravirt system ever shows up that does *not* have the | |
1543 | * ESPFIX issue, we can change this. | |
1544 | */ | |
1545 | #ifdef CONFIG_X86_32 | |
1546 | set_cpu_bug(c, X86_BUG_ESPFIX); | |
1547 | #endif | |
1548 | } | |
1549 | ||
1550 | /* | |
1551 | * Validate that ACPI/mptables have the same information about the | |
1552 | * effective APIC id and update the package map. | |
1553 | */ | |
1554 | static void validate_apic_and_package_id(struct cpuinfo_x86 *c) | |
1555 | { | |
1556 | #ifdef CONFIG_SMP | |
1557 | unsigned int apicid, cpu = smp_processor_id(); | |
1558 | ||
1559 | apicid = apic->cpu_present_to_apicid(cpu); | |
1560 | ||
1561 | if (apicid != c->apicid) { | |
1562 | pr_err(FW_BUG "CPU%u: APIC id mismatch. Firmware: %x APIC: %x\n", | |
1563 | cpu, apicid, c->initial_apicid); | |
1564 | } | |
1565 | BUG_ON(topology_update_package_map(c->phys_proc_id, cpu)); | |
1566 | BUG_ON(topology_update_die_map(c->cpu_die_id, cpu)); | |
1567 | #else | |
1568 | c->logical_proc_id = 0; | |
1569 | #endif | |
1570 | } | |
1571 | ||
1572 | /* | |
1573 | * This does the hard work of actually picking apart the CPU stuff... | |
1574 | */ | |
1575 | static void identify_cpu(struct cpuinfo_x86 *c) | |
1576 | { | |
1577 | int i; | |
1578 | ||
1579 | c->loops_per_jiffy = loops_per_jiffy; | |
1580 | c->x86_cache_size = 0; | |
1581 | c->x86_vendor = X86_VENDOR_UNKNOWN; | |
1582 | c->x86_model = c->x86_stepping = 0; /* So far unknown... */ | |
1583 | c->x86_vendor_id[0] = '\0'; /* Unset */ | |
1584 | c->x86_model_id[0] = '\0'; /* Unset */ | |
1585 | c->x86_max_cores = 1; | |
1586 | c->x86_coreid_bits = 0; | |
1587 | c->cu_id = 0xff; | |
1588 | #ifdef CONFIG_X86_64 | |
1589 | c->x86_clflush_size = 64; | |
1590 | c->x86_phys_bits = 36; | |
1591 | c->x86_virt_bits = 48; | |
1592 | #else | |
1593 | c->cpuid_level = -1; /* CPUID not detected */ | |
1594 | c->x86_clflush_size = 32; | |
1595 | c->x86_phys_bits = 32; | |
1596 | c->x86_virt_bits = 32; | |
1597 | #endif | |
1598 | c->x86_cache_alignment = c->x86_clflush_size; | |
1599 | memset(&c->x86_capability, 0, sizeof(c->x86_capability)); | |
1600 | #ifdef CONFIG_X86_VMX_FEATURE_NAMES | |
1601 | memset(&c->vmx_capability, 0, sizeof(c->vmx_capability)); | |
1602 | #endif | |
1603 | ||
1604 | generic_identify(c); | |
1605 | ||
1606 | if (this_cpu->c_identify) | |
1607 | this_cpu->c_identify(c); | |
1608 | ||
1609 | /* Clear/Set all flags overridden by options, after probe */ | |
1610 | apply_forced_caps(c); | |
1611 | ||
1612 | #ifdef CONFIG_X86_64 | |
1613 | c->apicid = apic->phys_pkg_id(c->initial_apicid, 0); | |
1614 | #endif | |
1615 | ||
1616 | /* | |
1617 | * Vendor-specific initialization. In this section we | |
1618 | * canonicalize the feature flags, meaning if there are | |
1619 | * features a certain CPU supports which CPUID doesn't | |
1620 | * tell us, CPUID claiming incorrect flags, or other bugs, | |
1621 | * we handle them here. | |
1622 | * | |
1623 | * At the end of this section, c->x86_capability better | |
1624 | * indicate the features this CPU genuinely supports! | |
1625 | */ | |
1626 | if (this_cpu->c_init) | |
1627 | this_cpu->c_init(c); | |
1628 | ||
1629 | /* Disable the PN if appropriate */ | |
1630 | squash_the_stupid_serial_number(c); | |
1631 | ||
1632 | /* Set up SMEP/SMAP/UMIP */ | |
1633 | setup_smep(c); | |
1634 | setup_smap(c); | |
1635 | setup_umip(c); | |
1636 | ||
1637 | /* Enable FSGSBASE instructions if available. */ | |
1638 | if (cpu_has(c, X86_FEATURE_FSGSBASE)) { | |
1639 | cr4_set_bits(X86_CR4_FSGSBASE); | |
1640 | elf_hwcap2 |= HWCAP2_FSGSBASE; | |
1641 | } | |
1642 | ||
1643 | /* | |
1644 | * The vendor-specific functions might have changed features. | |
1645 | * Now we do "generic changes." | |
1646 | */ | |
1647 | ||
1648 | /* Filter out anything that depends on CPUID levels we don't have */ | |
1649 | filter_cpuid_features(c, true); | |
1650 | ||
1651 | /* If the model name is still unset, do table lookup. */ | |
1652 | if (!c->x86_model_id[0]) { | |
1653 | const char *p; | |
1654 | p = table_lookup_model(c); | |
1655 | if (p) | |
1656 | strcpy(c->x86_model_id, p); | |
1657 | else | |
1658 | /* Last resort... */ | |
1659 | sprintf(c->x86_model_id, "%02x/%02x", | |
1660 | c->x86, c->x86_model); | |
1661 | } | |
1662 | ||
1663 | #ifdef CONFIG_X86_64 | |
1664 | detect_ht(c); | |
1665 | #endif | |
1666 | ||
1667 | x86_init_rdrand(c); | |
1668 | setup_pku(c); | |
1669 | ||
1670 | /* | |
1671 | * Clear/Set all flags overridden by options, need do it | |
1672 | * before following smp all cpus cap AND. | |
1673 | */ | |
1674 | apply_forced_caps(c); | |
1675 | ||
1676 | /* | |
1677 | * On SMP, boot_cpu_data holds the common feature set between | |
1678 | * all CPUs; so make sure that we indicate which features are | |
1679 | * common between the CPUs. The first time this routine gets | |
1680 | * executed, c == &boot_cpu_data. | |
1681 | */ | |
1682 | if (c != &boot_cpu_data) { | |
1683 | /* AND the already accumulated flags with these */ | |
1684 | for (i = 0; i < NCAPINTS; i++) | |
1685 | boot_cpu_data.x86_capability[i] &= c->x86_capability[i]; | |
1686 | ||
1687 | /* OR, i.e. replicate the bug flags */ | |
1688 | for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++) | |
1689 | c->x86_capability[i] |= boot_cpu_data.x86_capability[i]; | |
1690 | } | |
1691 | ||
1692 | /* Init Machine Check Exception if available. */ | |
1693 | mcheck_cpu_init(c); | |
1694 | ||
1695 | select_idle_routine(c); | |
1696 | ||
1697 | #ifdef CONFIG_NUMA | |
1698 | numa_add_cpu(smp_processor_id()); | |
1699 | #endif | |
1700 | } | |
1701 | ||
1702 | /* | |
1703 | * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions | |
1704 | * on 32-bit kernels: | |
1705 | */ | |
1706 | #ifdef CONFIG_X86_32 | |
1707 | void enable_sep_cpu(void) | |
1708 | { | |
1709 | struct tss_struct *tss; | |
1710 | int cpu; | |
1711 | ||
1712 | if (!boot_cpu_has(X86_FEATURE_SEP)) | |
1713 | return; | |
1714 | ||
1715 | cpu = get_cpu(); | |
1716 | tss = &per_cpu(cpu_tss_rw, cpu); | |
1717 | ||
1718 | /* | |
1719 | * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field -- | |
1720 | * see the big comment in struct x86_hw_tss's definition. | |
1721 | */ | |
1722 | ||
1723 | tss->x86_tss.ss1 = __KERNEL_CS; | |
1724 | wrmsr(MSR_IA32_SYSENTER_CS, tss->x86_tss.ss1, 0); | |
1725 | wrmsr(MSR_IA32_SYSENTER_ESP, (unsigned long)(cpu_entry_stack(cpu) + 1), 0); | |
1726 | wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long)entry_SYSENTER_32, 0); | |
1727 | ||
1728 | put_cpu(); | |
1729 | } | |
1730 | #endif | |
1731 | ||
1732 | void __init identify_boot_cpu(void) | |
1733 | { | |
1734 | identify_cpu(&boot_cpu_data); | |
1735 | #ifdef CONFIG_X86_32 | |
1736 | sysenter_setup(); | |
1737 | enable_sep_cpu(); | |
1738 | #endif | |
1739 | cpu_detect_tlb(&boot_cpu_data); | |
1740 | setup_cr_pinning(); | |
1741 | ||
1742 | tsx_init(); | |
1743 | } | |
1744 | ||
1745 | void identify_secondary_cpu(struct cpuinfo_x86 *c) | |
1746 | { | |
1747 | BUG_ON(c == &boot_cpu_data); | |
1748 | identify_cpu(c); | |
1749 | #ifdef CONFIG_X86_32 | |
1750 | enable_sep_cpu(); | |
1751 | #endif | |
1752 | mtrr_ap_init(); | |
1753 | validate_apic_and_package_id(c); | |
1754 | x86_spec_ctrl_setup_ap(); | |
1755 | update_srbds_msr(); | |
1756 | ||
1757 | tsx_ap_init(); | |
1758 | } | |
1759 | ||
1760 | static __init int setup_noclflush(char *arg) | |
1761 | { | |
1762 | setup_clear_cpu_cap(X86_FEATURE_CLFLUSH); | |
1763 | setup_clear_cpu_cap(X86_FEATURE_CLFLUSHOPT); | |
1764 | return 1; | |
1765 | } | |
1766 | __setup("noclflush", setup_noclflush); | |
1767 | ||
1768 | void print_cpu_info(struct cpuinfo_x86 *c) | |
1769 | { | |
1770 | const char *vendor = NULL; | |
1771 | ||
1772 | if (c->x86_vendor < X86_VENDOR_NUM) { | |
1773 | vendor = this_cpu->c_vendor; | |
1774 | } else { | |
1775 | if (c->cpuid_level >= 0) | |
1776 | vendor = c->x86_vendor_id; | |
1777 | } | |
1778 | ||
1779 | if (vendor && !strstr(c->x86_model_id, vendor)) | |
1780 | pr_cont("%s ", vendor); | |
1781 | ||
1782 | if (c->x86_model_id[0]) | |
1783 | pr_cont("%s", c->x86_model_id); | |
1784 | else | |
1785 | pr_cont("%d86", c->x86); | |
1786 | ||
1787 | pr_cont(" (family: 0x%x, model: 0x%x", c->x86, c->x86_model); | |
1788 | ||
1789 | if (c->x86_stepping || c->cpuid_level >= 0) | |
1790 | pr_cont(", stepping: 0x%x)\n", c->x86_stepping); | |
1791 | else | |
1792 | pr_cont(")\n"); | |
1793 | } | |
1794 | ||
1795 | /* | |
1796 | * clearcpuid= was already parsed in cpu_parse_early_param(). This dummy | |
1797 | * function prevents it from becoming an environment variable for init. | |
1798 | */ | |
1799 | static __init int setup_clearcpuid(char *arg) | |
1800 | { | |
1801 | return 1; | |
1802 | } | |
1803 | __setup("clearcpuid=", setup_clearcpuid); | |
1804 | ||
1805 | #ifdef CONFIG_X86_64 | |
1806 | DEFINE_PER_CPU_FIRST(struct fixed_percpu_data, | |
1807 | fixed_percpu_data) __aligned(PAGE_SIZE) __visible; | |
1808 | EXPORT_PER_CPU_SYMBOL_GPL(fixed_percpu_data); | |
1809 | ||
1810 | /* | |
1811 | * The following percpu variables are hot. Align current_task to | |
1812 | * cacheline size such that they fall in the same cacheline. | |
1813 | */ | |
1814 | DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned = | |
1815 | &init_task; | |
1816 | EXPORT_PER_CPU_SYMBOL(current_task); | |
1817 | ||
1818 | DEFINE_PER_CPU(void *, hardirq_stack_ptr); | |
1819 | DEFINE_PER_CPU(bool, hardirq_stack_inuse); | |
1820 | ||
1821 | DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT; | |
1822 | EXPORT_PER_CPU_SYMBOL(__preempt_count); | |
1823 | ||
1824 | DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack) = TOP_OF_INIT_STACK; | |
1825 | ||
1826 | /* May not be marked __init: used by software suspend */ | |
1827 | void syscall_init(void) | |
1828 | { | |
1829 | wrmsr(MSR_STAR, 0, (__USER32_CS << 16) | __KERNEL_CS); | |
1830 | wrmsrl(MSR_LSTAR, (unsigned long)entry_SYSCALL_64); | |
1831 | ||
1832 | #ifdef CONFIG_IA32_EMULATION | |
1833 | wrmsrl(MSR_CSTAR, (unsigned long)entry_SYSCALL_compat); | |
1834 | /* | |
1835 | * This only works on Intel CPUs. | |
1836 | * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP. | |
1837 | * This does not cause SYSENTER to jump to the wrong location, because | |
1838 | * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit). | |
1839 | */ | |
1840 | wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS); | |
1841 | wrmsrl_safe(MSR_IA32_SYSENTER_ESP, | |
1842 | (unsigned long)(cpu_entry_stack(smp_processor_id()) + 1)); | |
1843 | wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)entry_SYSENTER_compat); | |
1844 | #else | |
1845 | wrmsrl(MSR_CSTAR, (unsigned long)ignore_sysret); | |
1846 | wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG); | |
1847 | wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL); | |
1848 | wrmsrl_safe(MSR_IA32_SYSENTER_EIP, 0ULL); | |
1849 | #endif | |
1850 | ||
1851 | /* | |
1852 | * Flags to clear on syscall; clear as much as possible | |
1853 | * to minimize user space-kernel interference. | |
1854 | */ | |
1855 | wrmsrl(MSR_SYSCALL_MASK, | |
1856 | X86_EFLAGS_CF|X86_EFLAGS_PF|X86_EFLAGS_AF| | |
1857 | X86_EFLAGS_ZF|X86_EFLAGS_SF|X86_EFLAGS_TF| | |
1858 | X86_EFLAGS_IF|X86_EFLAGS_DF|X86_EFLAGS_OF| | |
1859 | X86_EFLAGS_IOPL|X86_EFLAGS_NT|X86_EFLAGS_RF| | |
1860 | X86_EFLAGS_AC|X86_EFLAGS_ID); | |
1861 | } | |
1862 | ||
1863 | #else /* CONFIG_X86_64 */ | |
1864 | ||
1865 | DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task; | |
1866 | EXPORT_PER_CPU_SYMBOL(current_task); | |
1867 | DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT; | |
1868 | EXPORT_PER_CPU_SYMBOL(__preempt_count); | |
1869 | ||
1870 | /* | |
1871 | * On x86_32, vm86 modifies tss.sp0, so sp0 isn't a reliable way to find | |
1872 | * the top of the kernel stack. Use an extra percpu variable to track the | |
1873 | * top of the kernel stack directly. | |
1874 | */ | |
1875 | DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack) = | |
1876 | (unsigned long)&init_thread_union + THREAD_SIZE; | |
1877 | EXPORT_PER_CPU_SYMBOL(cpu_current_top_of_stack); | |
1878 | ||
1879 | #ifdef CONFIG_STACKPROTECTOR | |
1880 | DEFINE_PER_CPU(unsigned long, __stack_chk_guard); | |
1881 | EXPORT_PER_CPU_SYMBOL(__stack_chk_guard); | |
1882 | #endif | |
1883 | ||
1884 | #endif /* CONFIG_X86_64 */ | |
1885 | ||
1886 | /* | |
1887 | * Clear all 6 debug registers: | |
1888 | */ | |
1889 | static void clear_all_debug_regs(void) | |
1890 | { | |
1891 | int i; | |
1892 | ||
1893 | for (i = 0; i < 8; i++) { | |
1894 | /* Ignore db4, db5 */ | |
1895 | if ((i == 4) || (i == 5)) | |
1896 | continue; | |
1897 | ||
1898 | set_debugreg(0, i); | |
1899 | } | |
1900 | } | |
1901 | ||
1902 | #ifdef CONFIG_KGDB | |
1903 | /* | |
1904 | * Restore debug regs if using kgdbwait and you have a kernel debugger | |
1905 | * connection established. | |
1906 | */ | |
1907 | static void dbg_restore_debug_regs(void) | |
1908 | { | |
1909 | if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break)) | |
1910 | arch_kgdb_ops.correct_hw_break(); | |
1911 | } | |
1912 | #else /* ! CONFIG_KGDB */ | |
1913 | #define dbg_restore_debug_regs() | |
1914 | #endif /* ! CONFIG_KGDB */ | |
1915 | ||
1916 | static void wait_for_master_cpu(int cpu) | |
1917 | { | |
1918 | #ifdef CONFIG_SMP | |
1919 | /* | |
1920 | * wait for ACK from master CPU before continuing | |
1921 | * with AP initialization | |
1922 | */ | |
1923 | WARN_ON(cpumask_test_and_set_cpu(cpu, cpu_initialized_mask)); | |
1924 | while (!cpumask_test_cpu(cpu, cpu_callout_mask)) | |
1925 | cpu_relax(); | |
1926 | #endif | |
1927 | } | |
1928 | ||
1929 | #ifdef CONFIG_X86_64 | |
1930 | static inline void setup_getcpu(int cpu) | |
1931 | { | |
1932 | unsigned long cpudata = vdso_encode_cpunode(cpu, early_cpu_to_node(cpu)); | |
1933 | struct desc_struct d = { }; | |
1934 | ||
1935 | if (boot_cpu_has(X86_FEATURE_RDTSCP) || boot_cpu_has(X86_FEATURE_RDPID)) | |
1936 | wrmsr(MSR_TSC_AUX, cpudata, 0); | |
1937 | ||
1938 | /* Store CPU and node number in limit. */ | |
1939 | d.limit0 = cpudata; | |
1940 | d.limit1 = cpudata >> 16; | |
1941 | ||
1942 | d.type = 5; /* RO data, expand down, accessed */ | |
1943 | d.dpl = 3; /* Visible to user code */ | |
1944 | d.s = 1; /* Not a system segment */ | |
1945 | d.p = 1; /* Present */ | |
1946 | d.d = 1; /* 32-bit */ | |
1947 | ||
1948 | write_gdt_entry(get_cpu_gdt_rw(cpu), GDT_ENTRY_CPUNODE, &d, DESCTYPE_S); | |
1949 | } | |
1950 | ||
1951 | static inline void ucode_cpu_init(int cpu) | |
1952 | { | |
1953 | if (cpu) | |
1954 | load_ucode_ap(); | |
1955 | } | |
1956 | ||
1957 | static inline void tss_setup_ist(struct tss_struct *tss) | |
1958 | { | |
1959 | /* Set up the per-CPU TSS IST stacks */ | |
1960 | tss->x86_tss.ist[IST_INDEX_DF] = __this_cpu_ist_top_va(DF); | |
1961 | tss->x86_tss.ist[IST_INDEX_NMI] = __this_cpu_ist_top_va(NMI); | |
1962 | tss->x86_tss.ist[IST_INDEX_DB] = __this_cpu_ist_top_va(DB); | |
1963 | tss->x86_tss.ist[IST_INDEX_MCE] = __this_cpu_ist_top_va(MCE); | |
1964 | /* Only mapped when SEV-ES is active */ | |
1965 | tss->x86_tss.ist[IST_INDEX_VC] = __this_cpu_ist_top_va(VC); | |
1966 | } | |
1967 | ||
1968 | #else /* CONFIG_X86_64 */ | |
1969 | ||
1970 | static inline void setup_getcpu(int cpu) { } | |
1971 | ||
1972 | static inline void ucode_cpu_init(int cpu) | |
1973 | { | |
1974 | show_ucode_info_early(); | |
1975 | } | |
1976 | ||
1977 | static inline void tss_setup_ist(struct tss_struct *tss) { } | |
1978 | ||
1979 | #endif /* !CONFIG_X86_64 */ | |
1980 | ||
1981 | static inline void tss_setup_io_bitmap(struct tss_struct *tss) | |
1982 | { | |
1983 | tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET_INVALID; | |
1984 | ||
1985 | #ifdef CONFIG_X86_IOPL_IOPERM | |
1986 | tss->io_bitmap.prev_max = 0; | |
1987 | tss->io_bitmap.prev_sequence = 0; | |
1988 | memset(tss->io_bitmap.bitmap, 0xff, sizeof(tss->io_bitmap.bitmap)); | |
1989 | /* | |
1990 | * Invalidate the extra array entry past the end of the all | |
1991 | * permission bitmap as required by the hardware. | |
1992 | */ | |
1993 | tss->io_bitmap.mapall[IO_BITMAP_LONGS] = ~0UL; | |
1994 | #endif | |
1995 | } | |
1996 | ||
1997 | /* | |
1998 | * Setup everything needed to handle exceptions from the IDT, including the IST | |
1999 | * exceptions which use paranoid_entry(). | |
2000 | */ | |
2001 | void cpu_init_exception_handling(void) | |
2002 | { | |
2003 | struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw); | |
2004 | int cpu = raw_smp_processor_id(); | |
2005 | ||
2006 | /* paranoid_entry() gets the CPU number from the GDT */ | |
2007 | setup_getcpu(cpu); | |
2008 | ||
2009 | /* IST vectors need TSS to be set up. */ | |
2010 | tss_setup_ist(tss); | |
2011 | tss_setup_io_bitmap(tss); | |
2012 | set_tss_desc(cpu, &get_cpu_entry_area(cpu)->tss.x86_tss); | |
2013 | ||
2014 | load_TR_desc(); | |
2015 | ||
2016 | /* Finally load the IDT */ | |
2017 | load_current_idt(); | |
2018 | } | |
2019 | ||
2020 | /* | |
2021 | * cpu_init() initializes state that is per-CPU. Some data is already | |
2022 | * initialized (naturally) in the bootstrap process, such as the GDT. We | |
2023 | * reload it nevertheless, this function acts as a 'CPU state barrier', | |
2024 | * nothing should get across. | |
2025 | */ | |
2026 | void cpu_init(void) | |
2027 | { | |
2028 | struct task_struct *cur = current; | |
2029 | int cpu = raw_smp_processor_id(); | |
2030 | ||
2031 | wait_for_master_cpu(cpu); | |
2032 | ||
2033 | ucode_cpu_init(cpu); | |
2034 | ||
2035 | #ifdef CONFIG_NUMA | |
2036 | if (this_cpu_read(numa_node) == 0 && | |
2037 | early_cpu_to_node(cpu) != NUMA_NO_NODE) | |
2038 | set_numa_node(early_cpu_to_node(cpu)); | |
2039 | #endif | |
2040 | pr_debug("Initializing CPU#%d\n", cpu); | |
2041 | ||
2042 | if (IS_ENABLED(CONFIG_X86_64) || cpu_feature_enabled(X86_FEATURE_VME) || | |
2043 | boot_cpu_has(X86_FEATURE_TSC) || boot_cpu_has(X86_FEATURE_DE)) | |
2044 | cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE); | |
2045 | ||
2046 | /* | |
2047 | * Initialize the per-CPU GDT with the boot GDT, | |
2048 | * and set up the GDT descriptor: | |
2049 | */ | |
2050 | switch_to_new_gdt(cpu); | |
2051 | ||
2052 | if (IS_ENABLED(CONFIG_X86_64)) { | |
2053 | loadsegment(fs, 0); | |
2054 | memset(cur->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8); | |
2055 | syscall_init(); | |
2056 | ||
2057 | wrmsrl(MSR_FS_BASE, 0); | |
2058 | wrmsrl(MSR_KERNEL_GS_BASE, 0); | |
2059 | barrier(); | |
2060 | ||
2061 | x2apic_setup(); | |
2062 | } | |
2063 | ||
2064 | mmgrab(&init_mm); | |
2065 | cur->active_mm = &init_mm; | |
2066 | BUG_ON(cur->mm); | |
2067 | initialize_tlbstate_and_flush(); | |
2068 | enter_lazy_tlb(&init_mm, cur); | |
2069 | ||
2070 | /* | |
2071 | * sp0 points to the entry trampoline stack regardless of what task | |
2072 | * is running. | |
2073 | */ | |
2074 | load_sp0((unsigned long)(cpu_entry_stack(cpu) + 1)); | |
2075 | ||
2076 | load_mm_ldt(&init_mm); | |
2077 | ||
2078 | clear_all_debug_regs(); | |
2079 | dbg_restore_debug_regs(); | |
2080 | ||
2081 | doublefault_init_cpu_tss(); | |
2082 | ||
2083 | fpu__init_cpu(); | |
2084 | ||
2085 | if (is_uv_system()) | |
2086 | uv_cpu_init(); | |
2087 | ||
2088 | load_fixmap_gdt(cpu); | |
2089 | } | |
2090 | ||
2091 | #ifdef CONFIG_SMP | |
2092 | void cpu_init_secondary(void) | |
2093 | { | |
2094 | /* | |
2095 | * Relies on the BP having set-up the IDT tables, which are loaded | |
2096 | * on this CPU in cpu_init_exception_handling(). | |
2097 | */ | |
2098 | cpu_init_exception_handling(); | |
2099 | cpu_init(); | |
2100 | } | |
2101 | #endif | |
2102 | ||
2103 | /* | |
2104 | * The microcode loader calls this upon late microcode load to recheck features, | |
2105 | * only when microcode has been updated. Caller holds microcode_mutex and CPU | |
2106 | * hotplug lock. | |
2107 | */ | |
2108 | void microcode_check(void) | |
2109 | { | |
2110 | struct cpuinfo_x86 info; | |
2111 | ||
2112 | perf_check_microcode(); | |
2113 | ||
2114 | /* Reload CPUID max function as it might've changed. */ | |
2115 | info.cpuid_level = cpuid_eax(0); | |
2116 | ||
2117 | /* | |
2118 | * Copy all capability leafs to pick up the synthetic ones so that | |
2119 | * memcmp() below doesn't fail on that. The ones coming from CPUID will | |
2120 | * get overwritten in get_cpu_cap(). | |
2121 | */ | |
2122 | memcpy(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability)); | |
2123 | ||
2124 | get_cpu_cap(&info); | |
2125 | ||
2126 | if (!memcmp(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability))) | |
2127 | return; | |
2128 | ||
2129 | pr_warn("x86/CPU: CPU features have changed after loading microcode, but might not take effect.\n"); | |
2130 | pr_warn("x86/CPU: Please consider either early loading through initrd/built-in or a potential BIOS update.\n"); | |
2131 | } | |
2132 | ||
2133 | /* | |
2134 | * Invoked from core CPU hotplug code after hotplug operations | |
2135 | */ | |
2136 | void arch_smt_update(void) | |
2137 | { | |
2138 | /* Handle the speculative execution misfeatures */ | |
2139 | cpu_bugs_smt_update(); | |
2140 | /* Check whether IPI broadcasting can be enabled */ | |
2141 | apic_smt_update(); | |
2142 | } |