]>
Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
2458e53f KS |
2 | /* cpu_feature_enabled() cannot be used this early */ |
3 | #define USE_EARLY_PGTABLE_L5 | |
4 | ||
57c8a661 | 5 | #include <linux/memblock.h> |
9766cdbc | 6 | #include <linux/linkage.h> |
f0fc4aff | 7 | #include <linux/bitops.h> |
9766cdbc | 8 | #include <linux/kernel.h> |
186f4360 | 9 | #include <linux/export.h> |
9766cdbc JSR |
10 | #include <linux/percpu.h> |
11 | #include <linux/string.h> | |
ee098e1a | 12 | #include <linux/ctype.h> |
1da177e4 | 13 | #include <linux/delay.h> |
68e21be2 | 14 | #include <linux/sched/mm.h> |
e6017571 | 15 | #include <linux/sched/clock.h> |
9164bb4a | 16 | #include <linux/sched/task.h> |
b47a3698 | 17 | #include <linux/sched/smt.h> |
9766cdbc | 18 | #include <linux/init.h> |
0f46efeb | 19 | #include <linux/kprobes.h> |
9766cdbc | 20 | #include <linux/kgdb.h> |
1da177e4 | 21 | #include <linux/smp.h> |
9766cdbc | 22 | #include <linux/io.h> |
b51ef52d | 23 | #include <linux/syscore_ops.h> |
65fddcfc | 24 | #include <linux/pgtable.h> |
9766cdbc | 25 | |
1ef5423a | 26 | #include <asm/cmdline.h> |
9766cdbc | 27 | #include <asm/stackprotector.h> |
cdd6c482 | 28 | #include <asm/perf_event.h> |
1da177e4 | 29 | #include <asm/mmu_context.h> |
dc4e0021 | 30 | #include <asm/doublefault.h> |
49d859d7 | 31 | #include <asm/archrandom.h> |
9766cdbc JSR |
32 | #include <asm/hypervisor.h> |
33 | #include <asm/processor.h> | |
1e02ce4c | 34 | #include <asm/tlbflush.h> |
f649e938 | 35 | #include <asm/debugreg.h> |
9766cdbc | 36 | #include <asm/sections.h> |
f40c3300 | 37 | #include <asm/vsyscall.h> |
8bdbd962 AC |
38 | #include <linux/topology.h> |
39 | #include <linux/cpumask.h> | |
60063497 | 40 | #include <linux/atomic.h> |
9766cdbc JSR |
41 | #include <asm/proto.h> |
42 | #include <asm/setup.h> | |
43 | #include <asm/apic.h> | |
44 | #include <asm/desc.h> | |
78f7f1e5 | 45 | #include <asm/fpu/internal.h> |
27b07da7 | 46 | #include <asm/mtrr.h> |
0274f955 | 47 | #include <asm/hwcap2.h> |
8bdbd962 | 48 | #include <linux/numa.h> |
0cd39f46 | 49 | #include <asm/numa.h> |
9766cdbc | 50 | #include <asm/asm.h> |
0f6ff2bc | 51 | #include <asm/bugs.h> |
9766cdbc | 52 | #include <asm/cpu.h> |
a03a3e28 | 53 | #include <asm/mce.h> |
9766cdbc | 54 | #include <asm/msr.h> |
eb243d1d | 55 | #include <asm/memtype.h> |
d288e1cf FY |
56 | #include <asm/microcode.h> |
57 | #include <asm/microcode_intel.h> | |
fec9434a DW |
58 | #include <asm/intel-family.h> |
59 | #include <asm/cpu_device_id.h> | |
bdbcdd48 | 60 | #include <asm/uv/uv.h> |
1da177e4 LT |
61 | |
62 | #include "cpu.h" | |
63 | ||
0274f955 GA |
64 | u32 elf_hwcap2 __read_mostly; |
65 | ||
c2d1cec1 | 66 | /* all of these masks are initialized in setup_cpu_local_masks() */ |
c2d1cec1 | 67 | cpumask_var_t cpu_initialized_mask; |
9766cdbc JSR |
68 | cpumask_var_t cpu_callout_mask; |
69 | cpumask_var_t cpu_callin_mask; | |
c2d1cec1 MT |
70 | |
71 | /* representing cpus for which sibling maps can be computed */ | |
72 | cpumask_var_t cpu_sibling_setup_mask; | |
73 | ||
f8b64d08 BP |
74 | /* Number of siblings per CPU package */ |
75 | int smp_num_siblings = 1; | |
76 | EXPORT_SYMBOL(smp_num_siblings); | |
77 | ||
78 | /* Last level cache ID of each logical CPU */ | |
79 | DEFINE_PER_CPU_READ_MOSTLY(u16, cpu_llc_id) = BAD_APICID; | |
80 | ||
2f2f52ba | 81 | /* correctly size the local cpu masks */ |
4369f1fb | 82 | void __init setup_cpu_local_masks(void) |
2f2f52ba BG |
83 | { |
84 | alloc_bootmem_cpumask_var(&cpu_initialized_mask); | |
85 | alloc_bootmem_cpumask_var(&cpu_callin_mask); | |
86 | alloc_bootmem_cpumask_var(&cpu_callout_mask); | |
87 | alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask); | |
88 | } | |
89 | ||
148f9bb8 | 90 | static void default_init(struct cpuinfo_x86 *c) |
e8055139 OZ |
91 | { |
92 | #ifdef CONFIG_X86_64 | |
27c13ece | 93 | cpu_detect_cache_sizes(c); |
e8055139 OZ |
94 | #else |
95 | /* Not much we can do here... */ | |
96 | /* Check if at least it has cpuid */ | |
97 | if (c->cpuid_level == -1) { | |
98 | /* No cpuid. It must be an ancient CPU */ | |
99 | if (c->x86 == 4) | |
100 | strcpy(c->x86_model_id, "486"); | |
101 | else if (c->x86 == 3) | |
102 | strcpy(c->x86_model_id, "386"); | |
103 | } | |
104 | #endif | |
105 | } | |
106 | ||
148f9bb8 | 107 | static const struct cpu_dev default_cpu = { |
e8055139 OZ |
108 | .c_init = default_init, |
109 | .c_vendor = "Unknown", | |
110 | .c_x86_vendor = X86_VENDOR_UNKNOWN, | |
111 | }; | |
112 | ||
148f9bb8 | 113 | static const struct cpu_dev *this_cpu = &default_cpu; |
0a488a53 | 114 | |
06deef89 | 115 | DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = { |
950ad7ff | 116 | #ifdef CONFIG_X86_64 |
06deef89 BG |
117 | /* |
118 | * We need valid kernel segments for data and code in long mode too | |
119 | * IRET will check the segment types kkeil 2000/10/28 | |
120 | * Also sysret mandates a special GDT layout | |
121 | * | |
9766cdbc | 122 | * TLS descriptors are currently at a different place compared to i386. |
06deef89 BG |
123 | * Hopefully nobody expects them at a fixed place (Wine?) |
124 | */ | |
1e5de182 AM |
125 | [GDT_ENTRY_KERNEL32_CS] = GDT_ENTRY_INIT(0xc09b, 0, 0xfffff), |
126 | [GDT_ENTRY_KERNEL_CS] = GDT_ENTRY_INIT(0xa09b, 0, 0xfffff), | |
127 | [GDT_ENTRY_KERNEL_DS] = GDT_ENTRY_INIT(0xc093, 0, 0xfffff), | |
128 | [GDT_ENTRY_DEFAULT_USER32_CS] = GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff), | |
129 | [GDT_ENTRY_DEFAULT_USER_DS] = GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff), | |
130 | [GDT_ENTRY_DEFAULT_USER_CS] = GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff), | |
950ad7ff | 131 | #else |
1e5de182 AM |
132 | [GDT_ENTRY_KERNEL_CS] = GDT_ENTRY_INIT(0xc09a, 0, 0xfffff), |
133 | [GDT_ENTRY_KERNEL_DS] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff), | |
134 | [GDT_ENTRY_DEFAULT_USER_CS] = GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff), | |
135 | [GDT_ENTRY_DEFAULT_USER_DS] = GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff), | |
bf504672 RR |
136 | /* |
137 | * Segments used for calling PnP BIOS have byte granularity. | |
138 | * They code segments and data segments have fixed 64k limits, | |
139 | * the transfer segment sizes are set at run time. | |
140 | */ | |
6842ef0e | 141 | /* 32-bit code */ |
1e5de182 | 142 | [GDT_ENTRY_PNPBIOS_CS32] = GDT_ENTRY_INIT(0x409a, 0, 0xffff), |
6842ef0e | 143 | /* 16-bit code */ |
1e5de182 | 144 | [GDT_ENTRY_PNPBIOS_CS16] = GDT_ENTRY_INIT(0x009a, 0, 0xffff), |
6842ef0e | 145 | /* 16-bit data */ |
1e5de182 | 146 | [GDT_ENTRY_PNPBIOS_DS] = GDT_ENTRY_INIT(0x0092, 0, 0xffff), |
6842ef0e | 147 | /* 16-bit data */ |
1e5de182 | 148 | [GDT_ENTRY_PNPBIOS_TS1] = GDT_ENTRY_INIT(0x0092, 0, 0), |
6842ef0e | 149 | /* 16-bit data */ |
1e5de182 | 150 | [GDT_ENTRY_PNPBIOS_TS2] = GDT_ENTRY_INIT(0x0092, 0, 0), |
bf504672 RR |
151 | /* |
152 | * The APM segments have byte granularity and their bases | |
153 | * are set at run time. All have 64k limits. | |
154 | */ | |
6842ef0e | 155 | /* 32-bit code */ |
1e5de182 | 156 | [GDT_ENTRY_APMBIOS_BASE] = GDT_ENTRY_INIT(0x409a, 0, 0xffff), |
bf504672 | 157 | /* 16-bit code */ |
1e5de182 | 158 | [GDT_ENTRY_APMBIOS_BASE+1] = GDT_ENTRY_INIT(0x009a, 0, 0xffff), |
6842ef0e | 159 | /* data */ |
72c4d853 | 160 | [GDT_ENTRY_APMBIOS_BASE+2] = GDT_ENTRY_INIT(0x4092, 0, 0xffff), |
bf504672 | 161 | |
1e5de182 AM |
162 | [GDT_ENTRY_ESPFIX_SS] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff), |
163 | [GDT_ENTRY_PERCPU] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff), | |
60a5317f | 164 | GDT_STACK_CANARY_INIT |
950ad7ff | 165 | #endif |
06deef89 | 166 | } }; |
7a61d35d | 167 | EXPORT_PER_CPU_SYMBOL_GPL(gdt_page); |
ae1ee11b | 168 | |
0790c9aa | 169 | #ifdef CONFIG_X86_64 |
c7ad5ad2 | 170 | static int __init x86_nopcid_setup(char *s) |
0790c9aa | 171 | { |
c7ad5ad2 AL |
172 | /* nopcid doesn't accept parameters */ |
173 | if (s) | |
174 | return -EINVAL; | |
0790c9aa AL |
175 | |
176 | /* do not emit a message if the feature is not present */ | |
177 | if (!boot_cpu_has(X86_FEATURE_PCID)) | |
c7ad5ad2 | 178 | return 0; |
0790c9aa AL |
179 | |
180 | setup_clear_cpu_cap(X86_FEATURE_PCID); | |
181 | pr_info("nopcid: PCID feature disabled\n"); | |
c7ad5ad2 | 182 | return 0; |
0790c9aa | 183 | } |
c7ad5ad2 | 184 | early_param("nopcid", x86_nopcid_setup); |
0790c9aa AL |
185 | #endif |
186 | ||
d12a72b8 AL |
187 | static int __init x86_noinvpcid_setup(char *s) |
188 | { | |
189 | /* noinvpcid doesn't accept parameters */ | |
190 | if (s) | |
191 | return -EINVAL; | |
192 | ||
193 | /* do not emit a message if the feature is not present */ | |
194 | if (!boot_cpu_has(X86_FEATURE_INVPCID)) | |
195 | return 0; | |
196 | ||
197 | setup_clear_cpu_cap(X86_FEATURE_INVPCID); | |
198 | pr_info("noinvpcid: INVPCID feature disabled\n"); | |
199 | return 0; | |
200 | } | |
201 | early_param("noinvpcid", x86_noinvpcid_setup); | |
202 | ||
ba51dced | 203 | #ifdef CONFIG_X86_32 |
148f9bb8 PG |
204 | static int cachesize_override = -1; |
205 | static int disable_x86_serial_nr = 1; | |
1da177e4 | 206 | |
0a488a53 YL |
207 | static int __init cachesize_setup(char *str) |
208 | { | |
209 | get_option(&str, &cachesize_override); | |
210 | return 1; | |
211 | } | |
212 | __setup("cachesize=", cachesize_setup); | |
213 | ||
0a488a53 YL |
214 | static int __init x86_sep_setup(char *s) |
215 | { | |
216 | setup_clear_cpu_cap(X86_FEATURE_SEP); | |
217 | return 1; | |
218 | } | |
219 | __setup("nosep", x86_sep_setup); | |
220 | ||
221 | /* Standard macro to see if a specific flag is changeable */ | |
222 | static inline int flag_is_changeable_p(u32 flag) | |
223 | { | |
224 | u32 f1, f2; | |
225 | ||
94f6bac1 KH |
226 | /* |
227 | * Cyrix and IDT cpus allow disabling of CPUID | |
228 | * so the code below may return different results | |
229 | * when it is executed before and after enabling | |
230 | * the CPUID. Add "volatile" to not allow gcc to | |
231 | * optimize the subsequent calls to this function. | |
232 | */ | |
0f3fa48a IM |
233 | asm volatile ("pushfl \n\t" |
234 | "pushfl \n\t" | |
235 | "popl %0 \n\t" | |
236 | "movl %0, %1 \n\t" | |
237 | "xorl %2, %0 \n\t" | |
238 | "pushl %0 \n\t" | |
239 | "popfl \n\t" | |
240 | "pushfl \n\t" | |
241 | "popl %0 \n\t" | |
242 | "popfl \n\t" | |
243 | ||
94f6bac1 KH |
244 | : "=&r" (f1), "=&r" (f2) |
245 | : "ir" (flag)); | |
0a488a53 YL |
246 | |
247 | return ((f1^f2) & flag) != 0; | |
248 | } | |
249 | ||
250 | /* Probe for the CPUID instruction */ | |
148f9bb8 | 251 | int have_cpuid_p(void) |
0a488a53 YL |
252 | { |
253 | return flag_is_changeable_p(X86_EFLAGS_ID); | |
254 | } | |
255 | ||
148f9bb8 | 256 | static void squash_the_stupid_serial_number(struct cpuinfo_x86 *c) |
0a488a53 | 257 | { |
0f3fa48a IM |
258 | unsigned long lo, hi; |
259 | ||
260 | if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr) | |
261 | return; | |
262 | ||
263 | /* Disable processor serial number: */ | |
264 | ||
265 | rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi); | |
266 | lo |= 0x200000; | |
267 | wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi); | |
268 | ||
1b74dde7 | 269 | pr_notice("CPU serial number disabled.\n"); |
0f3fa48a IM |
270 | clear_cpu_cap(c, X86_FEATURE_PN); |
271 | ||
272 | /* Disabling the serial number may affect the cpuid level */ | |
273 | c->cpuid_level = cpuid_eax(0); | |
0a488a53 YL |
274 | } |
275 | ||
276 | static int __init x86_serial_nr_setup(char *s) | |
277 | { | |
278 | disable_x86_serial_nr = 0; | |
279 | return 1; | |
280 | } | |
281 | __setup("serialnumber", x86_serial_nr_setup); | |
ba51dced | 282 | #else |
102bbe3a YL |
283 | static inline int flag_is_changeable_p(u32 flag) |
284 | { | |
285 | return 1; | |
286 | } | |
102bbe3a YL |
287 | static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c) |
288 | { | |
289 | } | |
ba51dced | 290 | #endif |
0a488a53 | 291 | |
de5397ad FY |
292 | static __init int setup_disable_smep(char *arg) |
293 | { | |
b2cc2a07 | 294 | setup_clear_cpu_cap(X86_FEATURE_SMEP); |
de5397ad FY |
295 | return 1; |
296 | } | |
297 | __setup("nosmep", setup_disable_smep); | |
298 | ||
b2cc2a07 | 299 | static __always_inline void setup_smep(struct cpuinfo_x86 *c) |
de5397ad | 300 | { |
b2cc2a07 | 301 | if (cpu_has(c, X86_FEATURE_SMEP)) |
375074cc | 302 | cr4_set_bits(X86_CR4_SMEP); |
de5397ad FY |
303 | } |
304 | ||
52b6179a PA |
305 | static __init int setup_disable_smap(char *arg) |
306 | { | |
b2cc2a07 | 307 | setup_clear_cpu_cap(X86_FEATURE_SMAP); |
52b6179a PA |
308 | return 1; |
309 | } | |
310 | __setup("nosmap", setup_disable_smap); | |
311 | ||
b2cc2a07 PA |
312 | static __always_inline void setup_smap(struct cpuinfo_x86 *c) |
313 | { | |
581b7f15 | 314 | unsigned long eflags = native_save_fl(); |
b2cc2a07 PA |
315 | |
316 | /* This should have been cleared long ago */ | |
b2cc2a07 PA |
317 | BUG_ON(eflags & X86_EFLAGS_AC); |
318 | ||
03bbd596 PA |
319 | if (cpu_has(c, X86_FEATURE_SMAP)) { |
320 | #ifdef CONFIG_X86_SMAP | |
375074cc | 321 | cr4_set_bits(X86_CR4_SMAP); |
03bbd596 | 322 | #else |
375074cc | 323 | cr4_clear_bits(X86_CR4_SMAP); |
03bbd596 PA |
324 | #endif |
325 | } | |
de5397ad FY |
326 | } |
327 | ||
aa35f896 RN |
328 | static __always_inline void setup_umip(struct cpuinfo_x86 *c) |
329 | { | |
330 | /* Check the boot processor, plus build option for UMIP. */ | |
331 | if (!cpu_feature_enabled(X86_FEATURE_UMIP)) | |
332 | goto out; | |
333 | ||
334 | /* Check the current processor's cpuid bits. */ | |
335 | if (!cpu_has(c, X86_FEATURE_UMIP)) | |
336 | goto out; | |
337 | ||
338 | cr4_set_bits(X86_CR4_UMIP); | |
339 | ||
438cbf88 | 340 | pr_info_once("x86/cpu: User Mode Instruction Prevention (UMIP) activated\n"); |
770c7755 | 341 | |
aa35f896 RN |
342 | return; |
343 | ||
344 | out: | |
345 | /* | |
346 | * Make sure UMIP is disabled in case it was enabled in a | |
347 | * previous boot (e.g., via kexec). | |
348 | */ | |
349 | cr4_clear_bits(X86_CR4_UMIP); | |
350 | } | |
351 | ||
a13b9d0b KC |
352 | /* These bits should not change their value after CPU init is finished. */ |
353 | static const unsigned long cr4_pinned_mask = | |
354 | X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_UMIP | X86_CR4_FSGSBASE; | |
7652ac92 TG |
355 | static DEFINE_STATIC_KEY_FALSE_RO(cr_pinning); |
356 | static unsigned long cr4_pinned_bits __ro_after_init; | |
357 | ||
358 | void native_write_cr0(unsigned long val) | |
359 | { | |
360 | unsigned long bits_missing = 0; | |
361 | ||
362 | set_register: | |
aa5cacdc | 363 | asm volatile("mov %0,%%cr0": "+r" (val) : : "memory"); |
7652ac92 TG |
364 | |
365 | if (static_branch_likely(&cr_pinning)) { | |
366 | if (unlikely((val & X86_CR0_WP) != X86_CR0_WP)) { | |
367 | bits_missing = X86_CR0_WP; | |
368 | val |= bits_missing; | |
369 | goto set_register; | |
370 | } | |
371 | /* Warn after we've set the missing bits. */ | |
372 | WARN_ONCE(bits_missing, "CR0 WP bit went missing!?\n"); | |
373 | } | |
374 | } | |
375 | EXPORT_SYMBOL(native_write_cr0); | |
376 | ||
377 | void native_write_cr4(unsigned long val) | |
378 | { | |
a13b9d0b | 379 | unsigned long bits_changed = 0; |
7652ac92 TG |
380 | |
381 | set_register: | |
aa5cacdc | 382 | asm volatile("mov %0,%%cr4": "+r" (val) : : "memory"); |
7652ac92 TG |
383 | |
384 | if (static_branch_likely(&cr_pinning)) { | |
a13b9d0b KC |
385 | if (unlikely((val & cr4_pinned_mask) != cr4_pinned_bits)) { |
386 | bits_changed = (val & cr4_pinned_mask) ^ cr4_pinned_bits; | |
387 | val = (val & ~cr4_pinned_mask) | cr4_pinned_bits; | |
7652ac92 TG |
388 | goto set_register; |
389 | } | |
a13b9d0b KC |
390 | /* Warn after we've corrected the changed bits. */ |
391 | WARN_ONCE(bits_changed, "pinned CR4 bits changed: 0x%lx!?\n", | |
392 | bits_changed); | |
7652ac92 TG |
393 | } |
394 | } | |
21953ee5 | 395 | #if IS_MODULE(CONFIG_LKDTM) |
d8f0b353 | 396 | EXPORT_SYMBOL_GPL(native_write_cr4); |
21953ee5 | 397 | #endif |
d8f0b353 TG |
398 | |
399 | void cr4_update_irqsoff(unsigned long set, unsigned long clear) | |
400 | { | |
401 | unsigned long newval, cr4 = this_cpu_read(cpu_tlbstate.cr4); | |
402 | ||
403 | lockdep_assert_irqs_disabled(); | |
404 | ||
405 | newval = (cr4 & ~clear) | set; | |
406 | if (newval != cr4) { | |
407 | this_cpu_write(cpu_tlbstate.cr4, newval); | |
408 | __write_cr4(newval); | |
409 | } | |
410 | } | |
411 | EXPORT_SYMBOL(cr4_update_irqsoff); | |
412 | ||
413 | /* Read the CR4 shadow. */ | |
414 | unsigned long cr4_read_shadow(void) | |
415 | { | |
416 | return this_cpu_read(cpu_tlbstate.cr4); | |
417 | } | |
418 | EXPORT_SYMBOL_GPL(cr4_read_shadow); | |
7652ac92 TG |
419 | |
420 | void cr4_init(void) | |
421 | { | |
422 | unsigned long cr4 = __read_cr4(); | |
423 | ||
424 | if (boot_cpu_has(X86_FEATURE_PCID)) | |
425 | cr4 |= X86_CR4_PCIDE; | |
426 | if (static_branch_likely(&cr_pinning)) | |
a13b9d0b | 427 | cr4 = (cr4 & ~cr4_pinned_mask) | cr4_pinned_bits; |
7652ac92 TG |
428 | |
429 | __write_cr4(cr4); | |
430 | ||
431 | /* Initialize cr4 shadow for this CPU. */ | |
432 | this_cpu_write(cpu_tlbstate.cr4, cr4); | |
433 | } | |
873d50d5 KC |
434 | |
435 | /* | |
436 | * Once CPU feature detection is finished (and boot params have been | |
437 | * parsed), record any of the sensitive CR bits that are set, and | |
438 | * enable CR pinning. | |
439 | */ | |
440 | static void __init setup_cr_pinning(void) | |
441 | { | |
a13b9d0b | 442 | cr4_pinned_bits = this_cpu_read(cpu_tlbstate.cr4) & cr4_pinned_mask; |
873d50d5 KC |
443 | static_key_enable(&cr_pinning.key); |
444 | } | |
445 | ||
b745cfba | 446 | static __init int x86_nofsgsbase_setup(char *arg) |
dd649bd0 | 447 | { |
b745cfba AL |
448 | /* Require an exact match without trailing characters. */ |
449 | if (strlen(arg)) | |
450 | return 0; | |
451 | ||
452 | /* Do not emit a message if the feature is not present. */ | |
453 | if (!boot_cpu_has(X86_FEATURE_FSGSBASE)) | |
454 | return 1; | |
455 | ||
456 | setup_clear_cpu_cap(X86_FEATURE_FSGSBASE); | |
457 | pr_info("FSGSBASE disabled via kernel command line\n"); | |
dd649bd0 AL |
458 | return 1; |
459 | } | |
b745cfba | 460 | __setup("nofsgsbase", x86_nofsgsbase_setup); |
dd649bd0 | 461 | |
06976945 DH |
462 | /* |
463 | * Protection Keys are not available in 32-bit mode. | |
464 | */ | |
465 | static bool pku_disabled; | |
466 | ||
467 | static __always_inline void setup_pku(struct cpuinfo_x86 *c) | |
468 | { | |
a5eff725 SAS |
469 | struct pkru_state *pk; |
470 | ||
e8df1a95 DH |
471 | /* check the boot processor, plus compile options for PKU: */ |
472 | if (!cpu_feature_enabled(X86_FEATURE_PKU)) | |
473 | return; | |
474 | /* checks the actual processor's cpuid bits: */ | |
06976945 DH |
475 | if (!cpu_has(c, X86_FEATURE_PKU)) |
476 | return; | |
477 | if (pku_disabled) | |
478 | return; | |
479 | ||
480 | cr4_set_bits(X86_CR4_PKE); | |
a5eff725 SAS |
481 | pk = get_xsave_addr(&init_fpstate.xsave, XFEATURE_PKRU); |
482 | if (pk) | |
483 | pk->pkru = init_pkru_value; | |
06976945 | 484 | /* |
d9f6e12f | 485 | * Setting X86_CR4_PKE will cause the X86_FEATURE_OSPKE |
06976945 DH |
486 | * cpuid bit to be set. We need to ensure that we |
487 | * update that bit in this CPU's "cpu_info". | |
488 | */ | |
735a6dd0 | 489 | set_cpu_cap(c, X86_FEATURE_OSPKE); |
06976945 DH |
490 | } |
491 | ||
492 | #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS | |
493 | static __init int setup_disable_pku(char *arg) | |
494 | { | |
495 | /* | |
496 | * Do not clear the X86_FEATURE_PKU bit. All of the | |
497 | * runtime checks are against OSPKE so clearing the | |
498 | * bit does nothing. | |
499 | * | |
500 | * This way, we will see "pku" in cpuinfo, but not | |
501 | * "ospke", which is exactly what we want. It shows | |
502 | * that the CPU has PKU, but the OS has not enabled it. | |
503 | * This happens to be exactly how a system would look | |
504 | * if we disabled the config option. | |
505 | */ | |
506 | pr_info("x86: 'nopku' specified, disabling Memory Protection Keys\n"); | |
507 | pku_disabled = true; | |
508 | return 1; | |
509 | } | |
510 | __setup("nopku", setup_disable_pku); | |
511 | #endif /* CONFIG_X86_64 */ | |
512 | ||
b38b0665 PA |
513 | /* |
514 | * Some CPU features depend on higher CPUID levels, which may not always | |
515 | * be available due to CPUID level capping or broken virtualization | |
516 | * software. Add those features to this table to auto-disable them. | |
517 | */ | |
518 | struct cpuid_dependent_feature { | |
519 | u32 feature; | |
520 | u32 level; | |
521 | }; | |
0f3fa48a | 522 | |
148f9bb8 | 523 | static const struct cpuid_dependent_feature |
b38b0665 PA |
524 | cpuid_dependent_features[] = { |
525 | { X86_FEATURE_MWAIT, 0x00000005 }, | |
526 | { X86_FEATURE_DCA, 0x00000009 }, | |
527 | { X86_FEATURE_XSAVE, 0x0000000d }, | |
528 | { 0, 0 } | |
529 | }; | |
530 | ||
148f9bb8 | 531 | static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn) |
b38b0665 PA |
532 | { |
533 | const struct cpuid_dependent_feature *df; | |
9766cdbc | 534 | |
b38b0665 | 535 | for (df = cpuid_dependent_features; df->feature; df++) { |
0f3fa48a IM |
536 | |
537 | if (!cpu_has(c, df->feature)) | |
538 | continue; | |
b38b0665 PA |
539 | /* |
540 | * Note: cpuid_level is set to -1 if unavailable, but | |
541 | * extended_extended_level is set to 0 if unavailable | |
542 | * and the legitimate extended levels are all negative | |
543 | * when signed; hence the weird messing around with | |
544 | * signs here... | |
545 | */ | |
0f3fa48a | 546 | if (!((s32)df->level < 0 ? |
f6db44df | 547 | (u32)df->level > (u32)c->extended_cpuid_level : |
0f3fa48a IM |
548 | (s32)df->level > (s32)c->cpuid_level)) |
549 | continue; | |
550 | ||
551 | clear_cpu_cap(c, df->feature); | |
552 | if (!warn) | |
553 | continue; | |
554 | ||
1b74dde7 CY |
555 | pr_warn("CPU: CPU feature " X86_CAP_FMT " disabled, no CPUID level 0x%x\n", |
556 | x86_cap_flag(df->feature), df->level); | |
b38b0665 | 557 | } |
f6db44df | 558 | } |
b38b0665 | 559 | |
102bbe3a YL |
560 | /* |
561 | * Naming convention should be: <Name> [(<Codename>)] | |
562 | * This table only is used unless init_<vendor>() below doesn't set it; | |
0f3fa48a IM |
563 | * in particular, if CPUID levels 0x80000002..4 are supported, this |
564 | * isn't used | |
102bbe3a YL |
565 | */ |
566 | ||
567 | /* Look up CPU names by table lookup. */ | |
148f9bb8 | 568 | static const char *table_lookup_model(struct cpuinfo_x86 *c) |
102bbe3a | 569 | { |
09dc68d9 JB |
570 | #ifdef CONFIG_X86_32 |
571 | const struct legacy_cpu_model_info *info; | |
102bbe3a YL |
572 | |
573 | if (c->x86_model >= 16) | |
574 | return NULL; /* Range check */ | |
575 | ||
576 | if (!this_cpu) | |
577 | return NULL; | |
578 | ||
09dc68d9 | 579 | info = this_cpu->legacy_models; |
102bbe3a | 580 | |
09dc68d9 | 581 | while (info->family) { |
102bbe3a YL |
582 | if (info->family == c->x86) |
583 | return info->model_names[c->x86_model]; | |
584 | info++; | |
585 | } | |
09dc68d9 | 586 | #endif |
102bbe3a YL |
587 | return NULL; /* Not found */ |
588 | } | |
589 | ||
f6a892dd FY |
590 | /* Aligned to unsigned long to avoid split lock in atomic bitmap ops */ |
591 | __u32 cpu_caps_cleared[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long)); | |
592 | __u32 cpu_caps_set[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long)); | |
7d851c8d | 593 | |
11e3a840 JF |
594 | void load_percpu_segment(int cpu) |
595 | { | |
596 | #ifdef CONFIG_X86_32 | |
597 | loadsegment(fs, __KERNEL_PERCPU); | |
598 | #else | |
45e876f7 | 599 | __loadsegment_simple(gs, 0); |
35060ed6 | 600 | wrmsrl(MSR_GS_BASE, cpu_kernelmode_gs_base(cpu)); |
11e3a840 | 601 | #endif |
60a5317f | 602 | load_stack_canary_segment(); |
11e3a840 JF |
603 | } |
604 | ||
72f5e08d AL |
605 | #ifdef CONFIG_X86_32 |
606 | /* The 32-bit entry code needs to find cpu_entry_area. */ | |
607 | DEFINE_PER_CPU(struct cpu_entry_area *, cpu_entry_area); | |
608 | #endif | |
609 | ||
45fc8757 TG |
610 | /* Load the original GDT from the per-cpu structure */ |
611 | void load_direct_gdt(int cpu) | |
612 | { | |
613 | struct desc_ptr gdt_descr; | |
614 | ||
615 | gdt_descr.address = (long)get_cpu_gdt_rw(cpu); | |
616 | gdt_descr.size = GDT_SIZE - 1; | |
617 | load_gdt(&gdt_descr); | |
618 | } | |
619 | EXPORT_SYMBOL_GPL(load_direct_gdt); | |
620 | ||
69218e47 TG |
621 | /* Load a fixmap remapping of the per-cpu GDT */ |
622 | void load_fixmap_gdt(int cpu) | |
623 | { | |
624 | struct desc_ptr gdt_descr; | |
625 | ||
626 | gdt_descr.address = (long)get_cpu_gdt_ro(cpu); | |
627 | gdt_descr.size = GDT_SIZE - 1; | |
628 | load_gdt(&gdt_descr); | |
629 | } | |
45fc8757 | 630 | EXPORT_SYMBOL_GPL(load_fixmap_gdt); |
69218e47 | 631 | |
0f3fa48a IM |
632 | /* |
633 | * Current gdt points %fs at the "master" per-cpu area: after this, | |
634 | * it's on the real one. | |
635 | */ | |
552be871 | 636 | void switch_to_new_gdt(int cpu) |
9d31d35b | 637 | { |
45fc8757 TG |
638 | /* Load the original GDT */ |
639 | load_direct_gdt(cpu); | |
2697fbd5 | 640 | /* Reload the per-cpu base */ |
11e3a840 | 641 | load_percpu_segment(cpu); |
9d31d35b YL |
642 | } |
643 | ||
148f9bb8 | 644 | static const struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {}; |
1da177e4 | 645 | |
148f9bb8 | 646 | static void get_model_name(struct cpuinfo_x86 *c) |
1da177e4 LT |
647 | { |
648 | unsigned int *v; | |
ee098e1a | 649 | char *p, *q, *s; |
1da177e4 | 650 | |
3da99c97 | 651 | if (c->extended_cpuid_level < 0x80000004) |
1b05d60d | 652 | return; |
1da177e4 | 653 | |
0f3fa48a | 654 | v = (unsigned int *)c->x86_model_id; |
1da177e4 LT |
655 | cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]); |
656 | cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]); | |
657 | cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]); | |
658 | c->x86_model_id[48] = 0; | |
659 | ||
ee098e1a BP |
660 | /* Trim whitespace */ |
661 | p = q = s = &c->x86_model_id[0]; | |
662 | ||
663 | while (*p == ' ') | |
664 | p++; | |
665 | ||
666 | while (*p) { | |
667 | /* Note the last non-whitespace index */ | |
668 | if (!isspace(*p)) | |
669 | s = q; | |
670 | ||
671 | *q++ = *p++; | |
672 | } | |
673 | ||
674 | *(s + 1) = '\0'; | |
1da177e4 LT |
675 | } |
676 | ||
9305bd6c | 677 | void detect_num_cpu_cores(struct cpuinfo_x86 *c) |
2cc61be6 DW |
678 | { |
679 | unsigned int eax, ebx, ecx, edx; | |
680 | ||
9305bd6c | 681 | c->x86_max_cores = 1; |
2cc61be6 | 682 | if (!IS_ENABLED(CONFIG_SMP) || c->cpuid_level < 4) |
9305bd6c | 683 | return; |
2cc61be6 DW |
684 | |
685 | cpuid_count(4, 0, &eax, &ebx, &ecx, &edx); | |
686 | if (eax & 0x1f) | |
9305bd6c | 687 | c->x86_max_cores = (eax >> 26) + 1; |
2cc61be6 DW |
688 | } |
689 | ||
148f9bb8 | 690 | void cpu_detect_cache_sizes(struct cpuinfo_x86 *c) |
1da177e4 | 691 | { |
9d31d35b | 692 | unsigned int n, dummy, ebx, ecx, edx, l2size; |
1da177e4 | 693 | |
3da99c97 | 694 | n = c->extended_cpuid_level; |
1da177e4 LT |
695 | |
696 | if (n >= 0x80000005) { | |
9d31d35b | 697 | cpuid(0x80000005, &dummy, &ebx, &ecx, &edx); |
9d31d35b | 698 | c->x86_cache_size = (ecx>>24) + (edx>>24); |
140fc727 YL |
699 | #ifdef CONFIG_X86_64 |
700 | /* On K8 L1 TLB is inclusive, so don't count it */ | |
701 | c->x86_tlbsize = 0; | |
702 | #endif | |
1da177e4 LT |
703 | } |
704 | ||
705 | if (n < 0x80000006) /* Some chips just has a large L1. */ | |
706 | return; | |
707 | ||
0a488a53 | 708 | cpuid(0x80000006, &dummy, &ebx, &ecx, &edx); |
1da177e4 | 709 | l2size = ecx >> 16; |
34048c9e | 710 | |
140fc727 YL |
711 | #ifdef CONFIG_X86_64 |
712 | c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff); | |
713 | #else | |
1da177e4 | 714 | /* do processor-specific cache resizing */ |
09dc68d9 JB |
715 | if (this_cpu->legacy_cache_size) |
716 | l2size = this_cpu->legacy_cache_size(c, l2size); | |
1da177e4 LT |
717 | |
718 | /* Allow user to override all this if necessary. */ | |
719 | if (cachesize_override != -1) | |
720 | l2size = cachesize_override; | |
721 | ||
34048c9e | 722 | if (l2size == 0) |
1da177e4 | 723 | return; /* Again, no L2 cache is possible */ |
140fc727 | 724 | #endif |
1da177e4 LT |
725 | |
726 | c->x86_cache_size = l2size; | |
1da177e4 LT |
727 | } |
728 | ||
e0ba94f1 AS |
729 | u16 __read_mostly tlb_lli_4k[NR_INFO]; |
730 | u16 __read_mostly tlb_lli_2m[NR_INFO]; | |
731 | u16 __read_mostly tlb_lli_4m[NR_INFO]; | |
732 | u16 __read_mostly tlb_lld_4k[NR_INFO]; | |
733 | u16 __read_mostly tlb_lld_2m[NR_INFO]; | |
734 | u16 __read_mostly tlb_lld_4m[NR_INFO]; | |
dd360393 | 735 | u16 __read_mostly tlb_lld_1g[NR_INFO]; |
e0ba94f1 | 736 | |
f94fe119 | 737 | static void cpu_detect_tlb(struct cpuinfo_x86 *c) |
e0ba94f1 AS |
738 | { |
739 | if (this_cpu->c_detect_tlb) | |
740 | this_cpu->c_detect_tlb(c); | |
741 | ||
f94fe119 | 742 | pr_info("Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n", |
e0ba94f1 | 743 | tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES], |
f94fe119 SH |
744 | tlb_lli_4m[ENTRIES]); |
745 | ||
746 | pr_info("Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n", | |
747 | tlb_lld_4k[ENTRIES], tlb_lld_2m[ENTRIES], | |
748 | tlb_lld_4m[ENTRIES], tlb_lld_1g[ENTRIES]); | |
e0ba94f1 AS |
749 | } |
750 | ||
545401f4 | 751 | int detect_ht_early(struct cpuinfo_x86 *c) |
1da177e4 | 752 | { |
c8e56d20 | 753 | #ifdef CONFIG_SMP |
0a488a53 | 754 | u32 eax, ebx, ecx, edx; |
1da177e4 | 755 | |
0a488a53 | 756 | if (!cpu_has(c, X86_FEATURE_HT)) |
545401f4 | 757 | return -1; |
1da177e4 | 758 | |
0a488a53 | 759 | if (cpu_has(c, X86_FEATURE_CMP_LEGACY)) |
545401f4 | 760 | return -1; |
1da177e4 | 761 | |
1cd78776 | 762 | if (cpu_has(c, X86_FEATURE_XTOPOLOGY)) |
545401f4 | 763 | return -1; |
1da177e4 | 764 | |
0a488a53 | 765 | cpuid(1, &eax, &ebx, &ecx, &edx); |
1da177e4 | 766 | |
9d31d35b | 767 | smp_num_siblings = (ebx & 0xff0000) >> 16; |
545401f4 | 768 | if (smp_num_siblings == 1) |
1b74dde7 | 769 | pr_info_once("CPU0: Hyper-Threading is disabled\n"); |
545401f4 TG |
770 | #endif |
771 | return 0; | |
772 | } | |
9d31d35b | 773 | |
545401f4 TG |
774 | void detect_ht(struct cpuinfo_x86 *c) |
775 | { | |
776 | #ifdef CONFIG_SMP | |
777 | int index_msb, core_bits; | |
55e6d279 | 778 | |
545401f4 | 779 | if (detect_ht_early(c) < 0) |
55e6d279 | 780 | return; |
9d31d35b | 781 | |
0f3fa48a IM |
782 | index_msb = get_count_order(smp_num_siblings); |
783 | c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb); | |
9d31d35b | 784 | |
0f3fa48a | 785 | smp_num_siblings = smp_num_siblings / c->x86_max_cores; |
9d31d35b | 786 | |
0f3fa48a | 787 | index_msb = get_count_order(smp_num_siblings); |
9d31d35b | 788 | |
0f3fa48a | 789 | core_bits = get_count_order(c->x86_max_cores); |
9d31d35b | 790 | |
0f3fa48a IM |
791 | c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) & |
792 | ((1 << core_bits) - 1); | |
9d31d35b | 793 | #endif |
97e4db7c | 794 | } |
1da177e4 | 795 | |
148f9bb8 | 796 | static void get_cpu_vendor(struct cpuinfo_x86 *c) |
1da177e4 LT |
797 | { |
798 | char *v = c->x86_vendor_id; | |
0f3fa48a | 799 | int i; |
1da177e4 LT |
800 | |
801 | for (i = 0; i < X86_VENDOR_NUM; i++) { | |
10a434fc YL |
802 | if (!cpu_devs[i]) |
803 | break; | |
804 | ||
805 | if (!strcmp(v, cpu_devs[i]->c_ident[0]) || | |
806 | (cpu_devs[i]->c_ident[1] && | |
807 | !strcmp(v, cpu_devs[i]->c_ident[1]))) { | |
0f3fa48a | 808 | |
10a434fc YL |
809 | this_cpu = cpu_devs[i]; |
810 | c->x86_vendor = this_cpu->c_x86_vendor; | |
811 | return; | |
1da177e4 LT |
812 | } |
813 | } | |
10a434fc | 814 | |
1b74dde7 CY |
815 | pr_err_once("CPU: vendor_id '%s' unknown, using generic init.\n" \ |
816 | "CPU: Your system may be unstable.\n", v); | |
10a434fc | 817 | |
fe38d855 CE |
818 | c->x86_vendor = X86_VENDOR_UNKNOWN; |
819 | this_cpu = &default_cpu; | |
1da177e4 LT |
820 | } |
821 | ||
148f9bb8 | 822 | void cpu_detect(struct cpuinfo_x86 *c) |
1da177e4 | 823 | { |
1da177e4 | 824 | /* Get vendor name */ |
4a148513 HH |
825 | cpuid(0x00000000, (unsigned int *)&c->cpuid_level, |
826 | (unsigned int *)&c->x86_vendor_id[0], | |
827 | (unsigned int *)&c->x86_vendor_id[8], | |
828 | (unsigned int *)&c->x86_vendor_id[4]); | |
1da177e4 | 829 | |
1da177e4 | 830 | c->x86 = 4; |
9d31d35b | 831 | /* Intel-defined flags: level 0x00000001 */ |
1da177e4 LT |
832 | if (c->cpuid_level >= 0x00000001) { |
833 | u32 junk, tfms, cap0, misc; | |
0f3fa48a | 834 | |
1da177e4 | 835 | cpuid(0x00000001, &tfms, &misc, &junk, &cap0); |
99f925ce BP |
836 | c->x86 = x86_family(tfms); |
837 | c->x86_model = x86_model(tfms); | |
b399151c | 838 | c->x86_stepping = x86_stepping(tfms); |
0f3fa48a | 839 | |
d4387bd3 | 840 | if (cap0 & (1<<19)) { |
d4387bd3 | 841 | c->x86_clflush_size = ((misc >> 8) & 0xff) * 8; |
9d31d35b | 842 | c->x86_cache_alignment = c->x86_clflush_size; |
d4387bd3 | 843 | } |
1da177e4 | 844 | } |
1da177e4 | 845 | } |
3da99c97 | 846 | |
8bf1ebca AL |
847 | static void apply_forced_caps(struct cpuinfo_x86 *c) |
848 | { | |
849 | int i; | |
850 | ||
6cbd2171 | 851 | for (i = 0; i < NCAPINTS + NBUGINTS; i++) { |
8bf1ebca AL |
852 | c->x86_capability[i] &= ~cpu_caps_cleared[i]; |
853 | c->x86_capability[i] |= cpu_caps_set[i]; | |
854 | } | |
855 | } | |
856 | ||
7fcae111 DW |
857 | static void init_speculation_control(struct cpuinfo_x86 *c) |
858 | { | |
859 | /* | |
860 | * The Intel SPEC_CTRL CPUID bit implies IBRS and IBPB support, | |
861 | * and they also have a different bit for STIBP support. Also, | |
862 | * a hypervisor might have set the individual AMD bits even on | |
863 | * Intel CPUs, for finer-grained selection of what's available. | |
7fcae111 DW |
864 | */ |
865 | if (cpu_has(c, X86_FEATURE_SPEC_CTRL)) { | |
866 | set_cpu_cap(c, X86_FEATURE_IBRS); | |
867 | set_cpu_cap(c, X86_FEATURE_IBPB); | |
7eb8956a | 868 | set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL); |
7fcae111 | 869 | } |
e7c587da | 870 | |
7fcae111 DW |
871 | if (cpu_has(c, X86_FEATURE_INTEL_STIBP)) |
872 | set_cpu_cap(c, X86_FEATURE_STIBP); | |
e7c587da | 873 | |
bc226f07 TL |
874 | if (cpu_has(c, X86_FEATURE_SPEC_CTRL_SSBD) || |
875 | cpu_has(c, X86_FEATURE_VIRT_SSBD)) | |
52817587 TG |
876 | set_cpu_cap(c, X86_FEATURE_SSBD); |
877 | ||
7eb8956a | 878 | if (cpu_has(c, X86_FEATURE_AMD_IBRS)) { |
e7c587da | 879 | set_cpu_cap(c, X86_FEATURE_IBRS); |
7eb8956a TG |
880 | set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL); |
881 | } | |
e7c587da BP |
882 | |
883 | if (cpu_has(c, X86_FEATURE_AMD_IBPB)) | |
884 | set_cpu_cap(c, X86_FEATURE_IBPB); | |
885 | ||
7eb8956a | 886 | if (cpu_has(c, X86_FEATURE_AMD_STIBP)) { |
e7c587da | 887 | set_cpu_cap(c, X86_FEATURE_STIBP); |
7eb8956a TG |
888 | set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL); |
889 | } | |
6ac2f49e KRW |
890 | |
891 | if (cpu_has(c, X86_FEATURE_AMD_SSBD)) { | |
892 | set_cpu_cap(c, X86_FEATURE_SSBD); | |
893 | set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL); | |
894 | clear_cpu_cap(c, X86_FEATURE_VIRT_SSBD); | |
895 | } | |
7fcae111 DW |
896 | } |
897 | ||
148f9bb8 | 898 | void get_cpu_cap(struct cpuinfo_x86 *c) |
093af8d7 | 899 | { |
39c06df4 | 900 | u32 eax, ebx, ecx, edx; |
093af8d7 | 901 | |
3da99c97 YL |
902 | /* Intel-defined flags: level 0x00000001 */ |
903 | if (c->cpuid_level >= 0x00000001) { | |
39c06df4 | 904 | cpuid(0x00000001, &eax, &ebx, &ecx, &edx); |
0f3fa48a | 905 | |
39c06df4 BP |
906 | c->x86_capability[CPUID_1_ECX] = ecx; |
907 | c->x86_capability[CPUID_1_EDX] = edx; | |
3da99c97 | 908 | } |
093af8d7 | 909 | |
3df8d920 AL |
910 | /* Thermal and Power Management Leaf: level 0x00000006 (eax) */ |
911 | if (c->cpuid_level >= 0x00000006) | |
912 | c->x86_capability[CPUID_6_EAX] = cpuid_eax(0x00000006); | |
913 | ||
bdc802dc PA |
914 | /* Additional Intel-defined flags: level 0x00000007 */ |
915 | if (c->cpuid_level >= 0x00000007) { | |
bdc802dc | 916 | cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx); |
39c06df4 | 917 | c->x86_capability[CPUID_7_0_EBX] = ebx; |
dfb4a70f | 918 | c->x86_capability[CPUID_7_ECX] = ecx; |
95ca0ee8 | 919 | c->x86_capability[CPUID_7_EDX] = edx; |
b302e4b1 FY |
920 | |
921 | /* Check valid sub-leaf index before accessing it */ | |
922 | if (eax >= 1) { | |
923 | cpuid_count(0x00000007, 1, &eax, &ebx, &ecx, &edx); | |
924 | c->x86_capability[CPUID_7_1_EAX] = eax; | |
925 | } | |
bdc802dc PA |
926 | } |
927 | ||
6229ad27 FY |
928 | /* Extended state features: level 0x0000000d */ |
929 | if (c->cpuid_level >= 0x0000000d) { | |
6229ad27 FY |
930 | cpuid_count(0x0000000d, 1, &eax, &ebx, &ecx, &edx); |
931 | ||
39c06df4 | 932 | c->x86_capability[CPUID_D_1_EAX] = eax; |
6229ad27 FY |
933 | } |
934 | ||
3da99c97 | 935 | /* AMD-defined flags: level 0x80000001 */ |
39c06df4 BP |
936 | eax = cpuid_eax(0x80000000); |
937 | c->extended_cpuid_level = eax; | |
938 | ||
939 | if ((eax & 0xffff0000) == 0x80000000) { | |
940 | if (eax >= 0x80000001) { | |
941 | cpuid(0x80000001, &eax, &ebx, &ecx, &edx); | |
0f3fa48a | 942 | |
39c06df4 BP |
943 | c->x86_capability[CPUID_8000_0001_ECX] = ecx; |
944 | c->x86_capability[CPUID_8000_0001_EDX] = edx; | |
093af8d7 | 945 | } |
093af8d7 | 946 | } |
093af8d7 | 947 | |
71faad43 YG |
948 | if (c->extended_cpuid_level >= 0x80000007) { |
949 | cpuid(0x80000007, &eax, &ebx, &ecx, &edx); | |
950 | ||
951 | c->x86_capability[CPUID_8000_0007_EBX] = ebx; | |
952 | c->x86_power = edx; | |
953 | } | |
954 | ||
c65732e4 TG |
955 | if (c->extended_cpuid_level >= 0x80000008) { |
956 | cpuid(0x80000008, &eax, &ebx, &ecx, &edx); | |
957 | c->x86_capability[CPUID_8000_0008_EBX] = ebx; | |
958 | } | |
959 | ||
2ccd71f1 | 960 | if (c->extended_cpuid_level >= 0x8000000a) |
39c06df4 | 961 | c->x86_capability[CPUID_8000_000A_EDX] = cpuid_edx(0x8000000a); |
093af8d7 | 962 | |
fb35d30f SC |
963 | if (c->extended_cpuid_level >= 0x8000001f) |
964 | c->x86_capability[CPUID_8000_001F_EAX] = cpuid_eax(0x8000001f); | |
965 | ||
1dedefd1 | 966 | init_scattered_cpuid_features(c); |
7fcae111 | 967 | init_speculation_control(c); |
60d34501 AL |
968 | |
969 | /* | |
970 | * Clear/Set all flags overridden by options, after probe. | |
971 | * This needs to happen each time we re-probe, which may happen | |
972 | * several times during CPU initialization. | |
973 | */ | |
974 | apply_forced_caps(c); | |
093af8d7 | 975 | } |
1da177e4 | 976 | |
405c018a | 977 | void get_cpu_address_sizes(struct cpuinfo_x86 *c) |
d94a155c KS |
978 | { |
979 | u32 eax, ebx, ecx, edx; | |
980 | ||
981 | if (c->extended_cpuid_level >= 0x80000008) { | |
982 | cpuid(0x80000008, &eax, &ebx, &ecx, &edx); | |
983 | ||
984 | c->x86_virt_bits = (eax >> 8) & 0xff; | |
985 | c->x86_phys_bits = eax & 0xff; | |
d94a155c KS |
986 | } |
987 | #ifdef CONFIG_X86_32 | |
988 | else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36)) | |
989 | c->x86_phys_bits = 36; | |
990 | #endif | |
cc51e542 | 991 | c->x86_cache_bits = c->x86_phys_bits; |
d94a155c KS |
992 | } |
993 | ||
148f9bb8 | 994 | static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c) |
aef93c8b YL |
995 | { |
996 | #ifdef CONFIG_X86_32 | |
997 | int i; | |
998 | ||
999 | /* | |
1000 | * First of all, decide if this is a 486 or higher | |
1001 | * It's a 486 if we can modify the AC flag | |
1002 | */ | |
1003 | if (flag_is_changeable_p(X86_EFLAGS_AC)) | |
1004 | c->x86 = 4; | |
1005 | else | |
1006 | c->x86 = 3; | |
1007 | ||
1008 | for (i = 0; i < X86_VENDOR_NUM; i++) | |
1009 | if (cpu_devs[i] && cpu_devs[i]->c_identify) { | |
1010 | c->x86_vendor_id[0] = 0; | |
1011 | cpu_devs[i]->c_identify(c); | |
1012 | if (c->x86_vendor_id[0]) { | |
1013 | get_cpu_vendor(c); | |
1014 | break; | |
1015 | } | |
1016 | } | |
1017 | #endif | |
1018 | } | |
1019 | ||
db4d30fb VT |
1020 | #define NO_SPECULATION BIT(0) |
1021 | #define NO_MELTDOWN BIT(1) | |
1022 | #define NO_SSB BIT(2) | |
1023 | #define NO_L1TF BIT(3) | |
1024 | #define NO_MDS BIT(4) | |
1025 | #define MSBDS_ONLY BIT(5) | |
1026 | #define NO_SWAPGS BIT(6) | |
1027 | #define NO_ITLB_MULTIHIT BIT(7) | |
1e41a766 | 1028 | #define NO_SPECTRE_V2 BIT(8) |
36ad3513 | 1029 | |
f6d502fc TG |
1030 | #define VULNWL(vendor, family, model, whitelist) \ |
1031 | X86_MATCH_VENDOR_FAM_MODEL(vendor, family, model, whitelist) | |
36ad3513 TG |
1032 | |
1033 | #define VULNWL_INTEL(model, whitelist) \ | |
1034 | VULNWL(INTEL, 6, INTEL_FAM6_##model, whitelist) | |
1035 | ||
1036 | #define VULNWL_AMD(family, whitelist) \ | |
1037 | VULNWL(AMD, family, X86_MODEL_ANY, whitelist) | |
1038 | ||
1039 | #define VULNWL_HYGON(family, whitelist) \ | |
1040 | VULNWL(HYGON, family, X86_MODEL_ANY, whitelist) | |
1041 | ||
1042 | static const __initconst struct x86_cpu_id cpu_vuln_whitelist[] = { | |
1043 | VULNWL(ANY, 4, X86_MODEL_ANY, NO_SPECULATION), | |
1044 | VULNWL(CENTAUR, 5, X86_MODEL_ANY, NO_SPECULATION), | |
1045 | VULNWL(INTEL, 5, X86_MODEL_ANY, NO_SPECULATION), | |
1046 | VULNWL(NSC, 5, X86_MODEL_ANY, NO_SPECULATION), | |
1047 | ||
ed5194c2 | 1048 | /* Intel Family 6 */ |
db4d30fb VT |
1049 | VULNWL_INTEL(ATOM_SALTWELL, NO_SPECULATION | NO_ITLB_MULTIHIT), |
1050 | VULNWL_INTEL(ATOM_SALTWELL_TABLET, NO_SPECULATION | NO_ITLB_MULTIHIT), | |
1051 | VULNWL_INTEL(ATOM_SALTWELL_MID, NO_SPECULATION | NO_ITLB_MULTIHIT), | |
1052 | VULNWL_INTEL(ATOM_BONNELL, NO_SPECULATION | NO_ITLB_MULTIHIT), | |
1053 | VULNWL_INTEL(ATOM_BONNELL_MID, NO_SPECULATION | NO_ITLB_MULTIHIT), | |
1054 | ||
1055 | VULNWL_INTEL(ATOM_SILVERMONT, NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT), | |
1056 | VULNWL_INTEL(ATOM_SILVERMONT_D, NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT), | |
1057 | VULNWL_INTEL(ATOM_SILVERMONT_MID, NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT), | |
1058 | VULNWL_INTEL(ATOM_AIRMONT, NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT), | |
1059 | VULNWL_INTEL(XEON_PHI_KNL, NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT), | |
1060 | VULNWL_INTEL(XEON_PHI_KNM, NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT), | |
36ad3513 TG |
1061 | |
1062 | VULNWL_INTEL(CORE_YONAH, NO_SSB), | |
1063 | ||
db4d30fb VT |
1064 | VULNWL_INTEL(ATOM_AIRMONT_MID, NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT), |
1065 | VULNWL_INTEL(ATOM_AIRMONT_NP, NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT), | |
36ad3513 | 1066 | |
db4d30fb VT |
1067 | VULNWL_INTEL(ATOM_GOLDMONT, NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT), |
1068 | VULNWL_INTEL(ATOM_GOLDMONT_D, NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT), | |
1069 | VULNWL_INTEL(ATOM_GOLDMONT_PLUS, NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT), | |
f36cf386 TG |
1070 | |
1071 | /* | |
1072 | * Technically, swapgs isn't serializing on AMD (despite it previously | |
1073 | * being documented as such in the APM). But according to AMD, %gs is | |
1074 | * updated non-speculatively, and the issuing of %gs-relative memory | |
1075 | * operands will be blocked until the %gs update completes, which is | |
1076 | * good enough for our purposes. | |
1077 | */ | |
ed5194c2 | 1078 | |
cad14885 PG |
1079 | VULNWL_INTEL(ATOM_TREMONT_D, NO_ITLB_MULTIHIT), |
1080 | ||
ed5194c2 | 1081 | /* AMD Family 0xf - 0x12 */ |
db4d30fb VT |
1082 | VULNWL_AMD(0x0f, NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT), |
1083 | VULNWL_AMD(0x10, NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT), | |
1084 | VULNWL_AMD(0x11, NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT), | |
1085 | VULNWL_AMD(0x12, NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT), | |
36ad3513 TG |
1086 | |
1087 | /* FAMILY_ANY must be last, otherwise 0x0f - 0x12 matches won't work */ | |
db4d30fb VT |
1088 | VULNWL_AMD(X86_FAMILY_ANY, NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT), |
1089 | VULNWL_HYGON(X86_FAMILY_ANY, NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT), | |
1e41a766 TW |
1090 | |
1091 | /* Zhaoxin Family 7 */ | |
a84de2fa TW |
1092 | VULNWL(CENTAUR, 7, X86_MODEL_ANY, NO_SPECTRE_V2 | NO_SWAPGS), |
1093 | VULNWL(ZHAOXIN, 7, X86_MODEL_ANY, NO_SPECTRE_V2 | NO_SWAPGS), | |
fec9434a DW |
1094 | {} |
1095 | }; | |
1096 | ||
7e5b3c26 MG |
1097 | #define VULNBL_INTEL_STEPPINGS(model, steppings, issues) \ |
1098 | X86_MATCH_VENDOR_FAM_MODEL_STEPPINGS_FEATURE(INTEL, 6, \ | |
1099 | INTEL_FAM6_##model, steppings, \ | |
1100 | X86_FEATURE_ANY, issues) | |
1101 | ||
1102 | #define SRBDS BIT(0) | |
1103 | ||
1104 | static const struct x86_cpu_id cpu_vuln_blacklist[] __initconst = { | |
1105 | VULNBL_INTEL_STEPPINGS(IVYBRIDGE, X86_STEPPING_ANY, SRBDS), | |
1106 | VULNBL_INTEL_STEPPINGS(HASWELL, X86_STEPPING_ANY, SRBDS), | |
1107 | VULNBL_INTEL_STEPPINGS(HASWELL_L, X86_STEPPING_ANY, SRBDS), | |
1108 | VULNBL_INTEL_STEPPINGS(HASWELL_G, X86_STEPPING_ANY, SRBDS), | |
1109 | VULNBL_INTEL_STEPPINGS(BROADWELL_G, X86_STEPPING_ANY, SRBDS), | |
1110 | VULNBL_INTEL_STEPPINGS(BROADWELL, X86_STEPPING_ANY, SRBDS), | |
1111 | VULNBL_INTEL_STEPPINGS(SKYLAKE_L, X86_STEPPING_ANY, SRBDS), | |
1112 | VULNBL_INTEL_STEPPINGS(SKYLAKE, X86_STEPPING_ANY, SRBDS), | |
1113 | VULNBL_INTEL_STEPPINGS(KABYLAKE_L, X86_STEPPINGS(0x0, 0xC), SRBDS), | |
1114 | VULNBL_INTEL_STEPPINGS(KABYLAKE, X86_STEPPINGS(0x0, 0xD), SRBDS), | |
1115 | {} | |
1116 | }; | |
1117 | ||
93920f61 | 1118 | static bool __init cpu_matches(const struct x86_cpu_id *table, unsigned long which) |
36ad3513 | 1119 | { |
93920f61 | 1120 | const struct x86_cpu_id *m = x86_match_cpu(table); |
c456442c | 1121 | |
36ad3513 TG |
1122 | return m && !!(m->driver_data & which); |
1123 | } | |
17dbca11 | 1124 | |
286836a7 | 1125 | u64 x86_read_arch_cap_msr(void) |
fec9434a DW |
1126 | { |
1127 | u64 ia32_cap = 0; | |
1128 | ||
286836a7 PG |
1129 | if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES)) |
1130 | rdmsrl(MSR_IA32_ARCH_CAPABILITIES, ia32_cap); | |
1131 | ||
1132 | return ia32_cap; | |
1133 | } | |
1134 | ||
1135 | static void __init cpu_set_bug_bits(struct cpuinfo_x86 *c) | |
1136 | { | |
1137 | u64 ia32_cap = x86_read_arch_cap_msr(); | |
1138 | ||
db4d30fb | 1139 | /* Set ITLB_MULTIHIT bug if cpu is not in the whitelist and not mitigated */ |
93920f61 MG |
1140 | if (!cpu_matches(cpu_vuln_whitelist, NO_ITLB_MULTIHIT) && |
1141 | !(ia32_cap & ARCH_CAP_PSCHANGE_MC_NO)) | |
db4d30fb VT |
1142 | setup_force_cpu_bug(X86_BUG_ITLB_MULTIHIT); |
1143 | ||
93920f61 | 1144 | if (cpu_matches(cpu_vuln_whitelist, NO_SPECULATION)) |
8ecc4979 DB |
1145 | return; |
1146 | ||
1147 | setup_force_cpu_bug(X86_BUG_SPECTRE_V1); | |
1e41a766 | 1148 | |
93920f61 | 1149 | if (!cpu_matches(cpu_vuln_whitelist, NO_SPECTRE_V2)) |
1e41a766 | 1150 | setup_force_cpu_bug(X86_BUG_SPECTRE_V2); |
8ecc4979 | 1151 | |
93920f61 MG |
1152 | if (!cpu_matches(cpu_vuln_whitelist, NO_SSB) && |
1153 | !(ia32_cap & ARCH_CAP_SSB_NO) && | |
24809860 | 1154 | !cpu_has(c, X86_FEATURE_AMD_SSB_NO)) |
c456442c KRW |
1155 | setup_force_cpu_bug(X86_BUG_SPEC_STORE_BYPASS); |
1156 | ||
706d5168 SP |
1157 | if (ia32_cap & ARCH_CAP_IBRS_ALL) |
1158 | setup_force_cpu_cap(X86_FEATURE_IBRS_ENHANCED); | |
1159 | ||
93920f61 MG |
1160 | if (!cpu_matches(cpu_vuln_whitelist, NO_MDS) && |
1161 | !(ia32_cap & ARCH_CAP_MDS_NO)) { | |
ed5194c2 | 1162 | setup_force_cpu_bug(X86_BUG_MDS); |
93920f61 | 1163 | if (cpu_matches(cpu_vuln_whitelist, MSBDS_ONLY)) |
e261f209 TG |
1164 | setup_force_cpu_bug(X86_BUG_MSBDS_ONLY); |
1165 | } | |
ed5194c2 | 1166 | |
93920f61 | 1167 | if (!cpu_matches(cpu_vuln_whitelist, NO_SWAPGS)) |
f36cf386 TG |
1168 | setup_force_cpu_bug(X86_BUG_SWAPGS); |
1169 | ||
1b42f017 PG |
1170 | /* |
1171 | * When the CPU is not mitigated for TAA (TAA_NO=0) set TAA bug when: | |
1172 | * - TSX is supported or | |
1173 | * - TSX_CTRL is present | |
1174 | * | |
1175 | * TSX_CTRL check is needed for cases when TSX could be disabled before | |
1176 | * the kernel boot e.g. kexec. | |
1177 | * TSX_CTRL check alone is not sufficient for cases when the microcode | |
1178 | * update is not present or running as guest that don't get TSX_CTRL. | |
1179 | */ | |
1180 | if (!(ia32_cap & ARCH_CAP_TAA_NO) && | |
1181 | (cpu_has(c, X86_FEATURE_RTM) || | |
1182 | (ia32_cap & ARCH_CAP_TSX_CTRL_MSR))) | |
1183 | setup_force_cpu_bug(X86_BUG_TAA); | |
1184 | ||
7e5b3c26 MG |
1185 | /* |
1186 | * SRBDS affects CPUs which support RDRAND or RDSEED and are listed | |
1187 | * in the vulnerability blacklist. | |
1188 | */ | |
1189 | if ((cpu_has(c, X86_FEATURE_RDRAND) || | |
1190 | cpu_has(c, X86_FEATURE_RDSEED)) && | |
1191 | cpu_matches(cpu_vuln_blacklist, SRBDS)) | |
1192 | setup_force_cpu_bug(X86_BUG_SRBDS); | |
1193 | ||
93920f61 | 1194 | if (cpu_matches(cpu_vuln_whitelist, NO_MELTDOWN)) |
4a28bfe3 | 1195 | return; |
fec9434a | 1196 | |
fec9434a DW |
1197 | /* Rogue Data Cache Load? No! */ |
1198 | if (ia32_cap & ARCH_CAP_RDCL_NO) | |
4a28bfe3 | 1199 | return; |
fec9434a | 1200 | |
4a28bfe3 | 1201 | setup_force_cpu_bug(X86_BUG_CPU_MELTDOWN); |
17dbca11 | 1202 | |
93920f61 | 1203 | if (cpu_matches(cpu_vuln_whitelist, NO_L1TF)) |
17dbca11 AK |
1204 | return; |
1205 | ||
1206 | setup_force_cpu_bug(X86_BUG_L1TF); | |
fec9434a DW |
1207 | } |
1208 | ||
8990cac6 PT |
1209 | /* |
1210 | * The NOPL instruction is supposed to exist on all CPUs of family >= 6; | |
1211 | * unfortunately, that's not true in practice because of early VIA | |
1212 | * chips and (more importantly) broken virtualizers that are not easy | |
1213 | * to detect. In the latter case it doesn't even *fail* reliably, so | |
1214 | * probing for it doesn't even work. Disable it completely on 32-bit | |
1215 | * unless we can find a reliable way to detect all the broken cases. | |
1216 | * Enable it explicitly on 64-bit for non-constant inputs of cpu_has(). | |
1217 | */ | |
9b3661cd | 1218 | static void detect_nopl(void) |
8990cac6 PT |
1219 | { |
1220 | #ifdef CONFIG_X86_32 | |
9b3661cd | 1221 | setup_clear_cpu_cap(X86_FEATURE_NOPL); |
8990cac6 | 1222 | #else |
9b3661cd | 1223 | setup_force_cpu_cap(X86_FEATURE_NOPL); |
8990cac6 PT |
1224 | #endif |
1225 | } | |
1226 | ||
1ef5423a MH |
1227 | /* |
1228 | * We parse cpu parameters early because fpu__init_system() is executed | |
1229 | * before parse_early_param(). | |
1230 | */ | |
1231 | static void __init cpu_parse_early_param(void) | |
1232 | { | |
1233 | char arg[128]; | |
1234 | char *argptr = arg; | |
1235 | int arglen, res, bit; | |
1236 | ||
1237 | #ifdef CONFIG_X86_32 | |
1238 | if (cmdline_find_option_bool(boot_command_line, "no387")) | |
1239 | #ifdef CONFIG_MATH_EMULATION | |
1240 | setup_clear_cpu_cap(X86_FEATURE_FPU); | |
1241 | #else | |
1242 | pr_err("Option 'no387' required CONFIG_MATH_EMULATION enabled.\n"); | |
1243 | #endif | |
1244 | ||
1245 | if (cmdline_find_option_bool(boot_command_line, "nofxsr")) | |
1246 | setup_clear_cpu_cap(X86_FEATURE_FXSR); | |
1247 | #endif | |
1248 | ||
1249 | if (cmdline_find_option_bool(boot_command_line, "noxsave")) | |
1250 | setup_clear_cpu_cap(X86_FEATURE_XSAVE); | |
1251 | ||
1252 | if (cmdline_find_option_bool(boot_command_line, "noxsaveopt")) | |
1253 | setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT); | |
1254 | ||
1255 | if (cmdline_find_option_bool(boot_command_line, "noxsaves")) | |
1256 | setup_clear_cpu_cap(X86_FEATURE_XSAVES); | |
1257 | ||
1258 | arglen = cmdline_find_option(boot_command_line, "clearcpuid", arg, sizeof(arg)); | |
1259 | if (arglen <= 0) | |
1260 | return; | |
1261 | ||
1262 | pr_info("Clearing CPUID bits:"); | |
1263 | do { | |
1264 | res = get_option(&argptr, &bit); | |
1265 | if (res == 0 || res == 3) | |
1266 | break; | |
1267 | ||
1268 | /* If the argument was too long, the last bit may be cut off */ | |
1269 | if (res == 1 && arglen >= sizeof(arg)) | |
1270 | break; | |
1271 | ||
1272 | if (bit >= 0 && bit < NCAPINTS * 32) { | |
1273 | pr_cont(" " X86_CAP_FMT, x86_cap_flag(bit)); | |
1274 | setup_clear_cpu_cap(bit); | |
1275 | } | |
1276 | } while (res == 2); | |
1277 | pr_cont("\n"); | |
1278 | } | |
1279 | ||
34048c9e PC |
1280 | /* |
1281 | * Do minimum CPU detection early. | |
1282 | * Fields really needed: vendor, cpuid_level, family, model, mask, | |
1283 | * cache alignment. | |
1284 | * The others are not touched to avoid unwanted side effects. | |
1285 | * | |
a1652bb8 JD |
1286 | * WARNING: this function is only called on the boot CPU. Don't add code |
1287 | * here that is supposed to run on all CPUs. | |
34048c9e | 1288 | */ |
3da99c97 | 1289 | static void __init early_identify_cpu(struct cpuinfo_x86 *c) |
d7cd5611 | 1290 | { |
6627d242 YL |
1291 | #ifdef CONFIG_X86_64 |
1292 | c->x86_clflush_size = 64; | |
13c6c532 JB |
1293 | c->x86_phys_bits = 36; |
1294 | c->x86_virt_bits = 48; | |
6627d242 | 1295 | #else |
d4387bd3 | 1296 | c->x86_clflush_size = 32; |
13c6c532 JB |
1297 | c->x86_phys_bits = 32; |
1298 | c->x86_virt_bits = 32; | |
6627d242 | 1299 | #endif |
0a488a53 | 1300 | c->x86_cache_alignment = c->x86_clflush_size; |
d7cd5611 | 1301 | |
0e96f31e | 1302 | memset(&c->x86_capability, 0, sizeof(c->x86_capability)); |
0a488a53 | 1303 | c->extended_cpuid_level = 0; |
d7cd5611 | 1304 | |
2893cc8f MW |
1305 | if (!have_cpuid_p()) |
1306 | identify_cpu_without_cpuid(c); | |
1307 | ||
aef93c8b | 1308 | /* cyrix could have cpuid enabled via c_identify()*/ |
05fb3c19 AL |
1309 | if (have_cpuid_p()) { |
1310 | cpu_detect(c); | |
1311 | get_cpu_vendor(c); | |
1312 | get_cpu_cap(c); | |
d94a155c | 1313 | get_cpu_address_sizes(c); |
78d1b296 | 1314 | setup_force_cpu_cap(X86_FEATURE_CPUID); |
1ef5423a | 1315 | cpu_parse_early_param(); |
d7cd5611 | 1316 | |
05fb3c19 AL |
1317 | if (this_cpu->c_early_init) |
1318 | this_cpu->c_early_init(c); | |
12cf105c | 1319 | |
05fb3c19 AL |
1320 | c->cpu_index = 0; |
1321 | filter_cpuid_features(c, false); | |
093af8d7 | 1322 | |
05fb3c19 AL |
1323 | if (this_cpu->c_bsp_init) |
1324 | this_cpu->c_bsp_init(c); | |
78d1b296 | 1325 | } else { |
78d1b296 | 1326 | setup_clear_cpu_cap(X86_FEATURE_CPUID); |
05fb3c19 | 1327 | } |
c3b83598 BP |
1328 | |
1329 | setup_force_cpu_cap(X86_FEATURE_ALWAYS); | |
a89f040f | 1330 | |
4a28bfe3 | 1331 | cpu_set_bug_bits(c); |
99c6fa25 | 1332 | |
ebb1064e | 1333 | sld_setup(c); |
6650cdd9 | 1334 | |
db52ef74 | 1335 | fpu__init_system(c); |
b8b7abae AL |
1336 | |
1337 | #ifdef CONFIG_X86_32 | |
1338 | /* | |
1339 | * Regardless of whether PCID is enumerated, the SDM says | |
1340 | * that it can't be enabled in 32-bit mode. | |
1341 | */ | |
1342 | setup_clear_cpu_cap(X86_FEATURE_PCID); | |
1343 | #endif | |
372fddf7 KS |
1344 | |
1345 | /* | |
1346 | * Later in the boot process pgtable_l5_enabled() relies on | |
1347 | * cpu_feature_enabled(X86_FEATURE_LA57). If 5-level paging is not | |
1348 | * enabled by this point we need to clear the feature bit to avoid | |
1349 | * false-positives at the later stage. | |
1350 | * | |
1351 | * pgtable_l5_enabled() can be false here for several reasons: | |
1352 | * - 5-level paging is disabled compile-time; | |
1353 | * - it's 32-bit kernel; | |
1354 | * - machine doesn't support 5-level paging; | |
1355 | * - user specified 'no5lvl' in kernel command line. | |
1356 | */ | |
1357 | if (!pgtable_l5_enabled()) | |
1358 | setup_clear_cpu_cap(X86_FEATURE_LA57); | |
8990cac6 | 1359 | |
9b3661cd | 1360 | detect_nopl(); |
d7cd5611 RR |
1361 | } |
1362 | ||
9d31d35b YL |
1363 | void __init early_cpu_init(void) |
1364 | { | |
02dde8b4 | 1365 | const struct cpu_dev *const *cdev; |
10a434fc YL |
1366 | int count = 0; |
1367 | ||
ac23f253 | 1368 | #ifdef CONFIG_PROCESSOR_SELECT |
1b74dde7 | 1369 | pr_info("KERNEL supported cpus:\n"); |
31c997ca IM |
1370 | #endif |
1371 | ||
10a434fc | 1372 | for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) { |
02dde8b4 | 1373 | const struct cpu_dev *cpudev = *cdev; |
9d31d35b | 1374 | |
10a434fc YL |
1375 | if (count >= X86_VENDOR_NUM) |
1376 | break; | |
1377 | cpu_devs[count] = cpudev; | |
1378 | count++; | |
1379 | ||
ac23f253 | 1380 | #ifdef CONFIG_PROCESSOR_SELECT |
31c997ca IM |
1381 | { |
1382 | unsigned int j; | |
1383 | ||
1384 | for (j = 0; j < 2; j++) { | |
1385 | if (!cpudev->c_ident[j]) | |
1386 | continue; | |
1b74dde7 | 1387 | pr_info(" %s %s\n", cpudev->c_vendor, |
31c997ca IM |
1388 | cpudev->c_ident[j]); |
1389 | } | |
10a434fc | 1390 | } |
0388423d | 1391 | #endif |
10a434fc | 1392 | } |
9d31d35b | 1393 | early_identify_cpu(&boot_cpu_data); |
d7cd5611 | 1394 | } |
093af8d7 | 1395 | |
7a5d6704 AL |
1396 | static void detect_null_seg_behavior(struct cpuinfo_x86 *c) |
1397 | { | |
1398 | #ifdef CONFIG_X86_64 | |
58a5aac5 | 1399 | /* |
7a5d6704 AL |
1400 | * Empirically, writing zero to a segment selector on AMD does |
1401 | * not clear the base, whereas writing zero to a segment | |
1402 | * selector on Intel does clear the base. Intel's behavior | |
1403 | * allows slightly faster context switches in the common case | |
1404 | * where GS is unused by the prev and next threads. | |
58a5aac5 | 1405 | * |
7a5d6704 | 1406 | * Since neither vendor documents this anywhere that I can see, |
d9f6e12f | 1407 | * detect it directly instead of hard-coding the choice by |
7a5d6704 AL |
1408 | * vendor. |
1409 | * | |
1410 | * I've designated AMD's behavior as the "bug" because it's | |
1411 | * counterintuitive and less friendly. | |
58a5aac5 | 1412 | */ |
7a5d6704 AL |
1413 | |
1414 | unsigned long old_base, tmp; | |
1415 | rdmsrl(MSR_FS_BASE, old_base); | |
1416 | wrmsrl(MSR_FS_BASE, 1); | |
1417 | loadsegment(fs, 0); | |
1418 | rdmsrl(MSR_FS_BASE, tmp); | |
1419 | if (tmp != 0) | |
1420 | set_cpu_bug(c, X86_BUG_NULL_SEG); | |
1421 | wrmsrl(MSR_FS_BASE, old_base); | |
366d4a43 | 1422 | #endif |
d7cd5611 RR |
1423 | } |
1424 | ||
148f9bb8 | 1425 | static void generic_identify(struct cpuinfo_x86 *c) |
1da177e4 | 1426 | { |
aef93c8b | 1427 | c->extended_cpuid_level = 0; |
1da177e4 | 1428 | |
3da99c97 | 1429 | if (!have_cpuid_p()) |
aef93c8b | 1430 | identify_cpu_without_cpuid(c); |
1d67953f | 1431 | |
aef93c8b | 1432 | /* cyrix could have cpuid enabled via c_identify()*/ |
a9853dd6 | 1433 | if (!have_cpuid_p()) |
aef93c8b | 1434 | return; |
1da177e4 | 1435 | |
3da99c97 | 1436 | cpu_detect(c); |
1da177e4 | 1437 | |
3da99c97 | 1438 | get_cpu_vendor(c); |
1da177e4 | 1439 | |
3da99c97 | 1440 | get_cpu_cap(c); |
1da177e4 | 1441 | |
d94a155c KS |
1442 | get_cpu_address_sizes(c); |
1443 | ||
3da99c97 YL |
1444 | if (c->cpuid_level >= 0x00000001) { |
1445 | c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF; | |
b89d3b3e | 1446 | #ifdef CONFIG_X86_32 |
c8e56d20 | 1447 | # ifdef CONFIG_SMP |
cb8cc442 | 1448 | c->apicid = apic->phys_pkg_id(c->initial_apicid, 0); |
b89d3b3e | 1449 | # else |
3da99c97 | 1450 | c->apicid = c->initial_apicid; |
b89d3b3e YL |
1451 | # endif |
1452 | #endif | |
b89d3b3e | 1453 | c->phys_proc_id = c->initial_apicid; |
3da99c97 | 1454 | } |
1da177e4 | 1455 | |
1b05d60d | 1456 | get_model_name(c); /* Default name */ |
1da177e4 | 1457 | |
7a5d6704 | 1458 | detect_null_seg_behavior(c); |
0230bb03 AL |
1459 | |
1460 | /* | |
1461 | * ESPFIX is a strange bug. All real CPUs have it. Paravirt | |
1462 | * systems that run Linux at CPL > 0 may or may not have the | |
1463 | * issue, but, even if they have the issue, there's absolutely | |
1464 | * nothing we can do about it because we can't use the real IRET | |
1465 | * instruction. | |
1466 | * | |
1467 | * NB: For the time being, only 32-bit kernels support | |
1468 | * X86_BUG_ESPFIX as such. 64-bit kernels directly choose | |
1469 | * whether to apply espfix using paravirt hooks. If any | |
1470 | * non-paravirt system ever shows up that does *not* have the | |
1471 | * ESPFIX issue, we can change this. | |
1472 | */ | |
1473 | #ifdef CONFIG_X86_32 | |
0230bb03 | 1474 | set_cpu_bug(c, X86_BUG_ESPFIX); |
0230bb03 | 1475 | #endif |
1da177e4 | 1476 | } |
1da177e4 | 1477 | |
d49597fd | 1478 | /* |
9d85eb91 TG |
1479 | * Validate that ACPI/mptables have the same information about the |
1480 | * effective APIC id and update the package map. | |
d49597fd | 1481 | */ |
9d85eb91 | 1482 | static void validate_apic_and_package_id(struct cpuinfo_x86 *c) |
d49597fd TG |
1483 | { |
1484 | #ifdef CONFIG_SMP | |
9d85eb91 | 1485 | unsigned int apicid, cpu = smp_processor_id(); |
d49597fd TG |
1486 | |
1487 | apicid = apic->cpu_present_to_apicid(cpu); | |
d49597fd | 1488 | |
9d85eb91 TG |
1489 | if (apicid != c->apicid) { |
1490 | pr_err(FW_BUG "CPU%u: APIC id mismatch. Firmware: %x APIC: %x\n", | |
d49597fd | 1491 | cpu, apicid, c->initial_apicid); |
d49597fd | 1492 | } |
9d85eb91 | 1493 | BUG_ON(topology_update_package_map(c->phys_proc_id, cpu)); |
212bf4fd | 1494 | BUG_ON(topology_update_die_map(c->cpu_die_id, cpu)); |
d49597fd TG |
1495 | #else |
1496 | c->logical_proc_id = 0; | |
1497 | #endif | |
1498 | } | |
1499 | ||
1da177e4 LT |
1500 | /* |
1501 | * This does the hard work of actually picking apart the CPU stuff... | |
1502 | */ | |
148f9bb8 | 1503 | static void identify_cpu(struct cpuinfo_x86 *c) |
1da177e4 LT |
1504 | { |
1505 | int i; | |
1506 | ||
1507 | c->loops_per_jiffy = loops_per_jiffy; | |
24dbc600 | 1508 | c->x86_cache_size = 0; |
1da177e4 | 1509 | c->x86_vendor = X86_VENDOR_UNKNOWN; |
b399151c | 1510 | c->x86_model = c->x86_stepping = 0; /* So far unknown... */ |
1da177e4 LT |
1511 | c->x86_vendor_id[0] = '\0'; /* Unset */ |
1512 | c->x86_model_id[0] = '\0'; /* Unset */ | |
94605eff | 1513 | c->x86_max_cores = 1; |
102bbe3a | 1514 | c->x86_coreid_bits = 0; |
79a8b9aa | 1515 | c->cu_id = 0xff; |
11fdd252 | 1516 | #ifdef CONFIG_X86_64 |
102bbe3a | 1517 | c->x86_clflush_size = 64; |
13c6c532 JB |
1518 | c->x86_phys_bits = 36; |
1519 | c->x86_virt_bits = 48; | |
102bbe3a YL |
1520 | #else |
1521 | c->cpuid_level = -1; /* CPUID not detected */ | |
770d132f | 1522 | c->x86_clflush_size = 32; |
13c6c532 JB |
1523 | c->x86_phys_bits = 32; |
1524 | c->x86_virt_bits = 32; | |
102bbe3a YL |
1525 | #endif |
1526 | c->x86_cache_alignment = c->x86_clflush_size; | |
0e96f31e | 1527 | memset(&c->x86_capability, 0, sizeof(c->x86_capability)); |
b47ce1fe SC |
1528 | #ifdef CONFIG_X86_VMX_FEATURE_NAMES |
1529 | memset(&c->vmx_capability, 0, sizeof(c->vmx_capability)); | |
1530 | #endif | |
1da177e4 | 1531 | |
1da177e4 LT |
1532 | generic_identify(c); |
1533 | ||
3898534d | 1534 | if (this_cpu->c_identify) |
1da177e4 LT |
1535 | this_cpu->c_identify(c); |
1536 | ||
6a6256f9 | 1537 | /* Clear/Set all flags overridden by options, after probe */ |
8bf1ebca | 1538 | apply_forced_caps(c); |
2759c328 | 1539 | |
102bbe3a | 1540 | #ifdef CONFIG_X86_64 |
cb8cc442 | 1541 | c->apicid = apic->phys_pkg_id(c->initial_apicid, 0); |
102bbe3a YL |
1542 | #endif |
1543 | ||
1da177e4 LT |
1544 | /* |
1545 | * Vendor-specific initialization. In this section we | |
1546 | * canonicalize the feature flags, meaning if there are | |
1547 | * features a certain CPU supports which CPUID doesn't | |
1548 | * tell us, CPUID claiming incorrect flags, or other bugs, | |
1549 | * we handle them here. | |
1550 | * | |
1551 | * At the end of this section, c->x86_capability better | |
1552 | * indicate the features this CPU genuinely supports! | |
1553 | */ | |
1554 | if (this_cpu->c_init) | |
1555 | this_cpu->c_init(c); | |
1556 | ||
1557 | /* Disable the PN if appropriate */ | |
1558 | squash_the_stupid_serial_number(c); | |
1559 | ||
aa35f896 | 1560 | /* Set up SMEP/SMAP/UMIP */ |
b2cc2a07 PA |
1561 | setup_smep(c); |
1562 | setup_smap(c); | |
aa35f896 | 1563 | setup_umip(c); |
b2cc2a07 | 1564 | |
dd649bd0 | 1565 | /* Enable FSGSBASE instructions if available. */ |
742c45c3 | 1566 | if (cpu_has(c, X86_FEATURE_FSGSBASE)) { |
b745cfba | 1567 | cr4_set_bits(X86_CR4_FSGSBASE); |
742c45c3 AK |
1568 | elf_hwcap2 |= HWCAP2_FSGSBASE; |
1569 | } | |
dd649bd0 | 1570 | |
1da177e4 | 1571 | /* |
0f3fa48a IM |
1572 | * The vendor-specific functions might have changed features. |
1573 | * Now we do "generic changes." | |
1da177e4 LT |
1574 | */ |
1575 | ||
b38b0665 PA |
1576 | /* Filter out anything that depends on CPUID levels we don't have */ |
1577 | filter_cpuid_features(c, true); | |
1578 | ||
1da177e4 | 1579 | /* If the model name is still unset, do table lookup. */ |
34048c9e | 1580 | if (!c->x86_model_id[0]) { |
02dde8b4 | 1581 | const char *p; |
1da177e4 | 1582 | p = table_lookup_model(c); |
34048c9e | 1583 | if (p) |
1da177e4 LT |
1584 | strcpy(c->x86_model_id, p); |
1585 | else | |
1586 | /* Last resort... */ | |
1587 | sprintf(c->x86_model_id, "%02x/%02x", | |
54a20f8c | 1588 | c->x86, c->x86_model); |
1da177e4 LT |
1589 | } |
1590 | ||
102bbe3a YL |
1591 | #ifdef CONFIG_X86_64 |
1592 | detect_ht(c); | |
1593 | #endif | |
1594 | ||
49d859d7 | 1595 | x86_init_rdrand(c); |
06976945 | 1596 | setup_pku(c); |
3e0c3737 YL |
1597 | |
1598 | /* | |
6a6256f9 | 1599 | * Clear/Set all flags overridden by options, need do it |
3e0c3737 YL |
1600 | * before following smp all cpus cap AND. |
1601 | */ | |
8bf1ebca | 1602 | apply_forced_caps(c); |
3e0c3737 | 1603 | |
1da177e4 LT |
1604 | /* |
1605 | * On SMP, boot_cpu_data holds the common feature set between | |
1606 | * all CPUs; so make sure that we indicate which features are | |
1607 | * common between the CPUs. The first time this routine gets | |
1608 | * executed, c == &boot_cpu_data. | |
1609 | */ | |
34048c9e | 1610 | if (c != &boot_cpu_data) { |
1da177e4 | 1611 | /* AND the already accumulated flags with these */ |
9d31d35b | 1612 | for (i = 0; i < NCAPINTS; i++) |
1da177e4 | 1613 | boot_cpu_data.x86_capability[i] &= c->x86_capability[i]; |
65fc985b BP |
1614 | |
1615 | /* OR, i.e. replicate the bug flags */ | |
1616 | for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++) | |
1617 | c->x86_capability[i] |= boot_cpu_data.x86_capability[i]; | |
1da177e4 LT |
1618 | } |
1619 | ||
1620 | /* Init Machine Check Exception if available. */ | |
5e09954a | 1621 | mcheck_cpu_init(c); |
30d432df AK |
1622 | |
1623 | select_idle_routine(c); | |
102bbe3a | 1624 | |
de2d9445 | 1625 | #ifdef CONFIG_NUMA |
102bbe3a YL |
1626 | numa_add_cpu(smp_processor_id()); |
1627 | #endif | |
a6c4e076 | 1628 | } |
31ab269a | 1629 | |
8b6c0ab1 IM |
1630 | /* |
1631 | * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions | |
1632 | * on 32-bit kernels: | |
1633 | */ | |
cfda7bb9 AL |
1634 | #ifdef CONFIG_X86_32 |
1635 | void enable_sep_cpu(void) | |
1636 | { | |
8b6c0ab1 IM |
1637 | struct tss_struct *tss; |
1638 | int cpu; | |
cfda7bb9 | 1639 | |
b3edfda4 BP |
1640 | if (!boot_cpu_has(X86_FEATURE_SEP)) |
1641 | return; | |
1642 | ||
8b6c0ab1 | 1643 | cpu = get_cpu(); |
c482feef | 1644 | tss = &per_cpu(cpu_tss_rw, cpu); |
8b6c0ab1 | 1645 | |
8b6c0ab1 | 1646 | /* |
cf9328cc AL |
1647 | * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field -- |
1648 | * see the big comment in struct x86_hw_tss's definition. | |
8b6c0ab1 | 1649 | */ |
cfda7bb9 AL |
1650 | |
1651 | tss->x86_tss.ss1 = __KERNEL_CS; | |
8b6c0ab1 | 1652 | wrmsr(MSR_IA32_SYSENTER_CS, tss->x86_tss.ss1, 0); |
4fe2d8b1 | 1653 | wrmsr(MSR_IA32_SYSENTER_ESP, (unsigned long)(cpu_entry_stack(cpu) + 1), 0); |
4c8cd0c5 | 1654 | wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long)entry_SYSENTER_32, 0); |
8b6c0ab1 | 1655 | |
cfda7bb9 AL |
1656 | put_cpu(); |
1657 | } | |
e04d645f GC |
1658 | #endif |
1659 | ||
a6c4e076 JF |
1660 | void __init identify_boot_cpu(void) |
1661 | { | |
1662 | identify_cpu(&boot_cpu_data); | |
102bbe3a | 1663 | #ifdef CONFIG_X86_32 |
a6c4e076 | 1664 | sysenter_setup(); |
6fe940d6 | 1665 | enable_sep_cpu(); |
102bbe3a | 1666 | #endif |
5b556332 | 1667 | cpu_detect_tlb(&boot_cpu_data); |
873d50d5 | 1668 | setup_cr_pinning(); |
95c5824f PG |
1669 | |
1670 | tsx_init(); | |
a6c4e076 | 1671 | } |
3b520b23 | 1672 | |
148f9bb8 | 1673 | void identify_secondary_cpu(struct cpuinfo_x86 *c) |
a6c4e076 JF |
1674 | { |
1675 | BUG_ON(c == &boot_cpu_data); | |
1676 | identify_cpu(c); | |
102bbe3a | 1677 | #ifdef CONFIG_X86_32 |
a6c4e076 | 1678 | enable_sep_cpu(); |
102bbe3a | 1679 | #endif |
a6c4e076 | 1680 | mtrr_ap_init(); |
9d85eb91 | 1681 | validate_apic_and_package_id(c); |
77243971 | 1682 | x86_spec_ctrl_setup_ap(); |
7e5b3c26 | 1683 | update_srbds_msr(); |
1da177e4 LT |
1684 | } |
1685 | ||
191679fd AK |
1686 | static __init int setup_noclflush(char *arg) |
1687 | { | |
840d2830 | 1688 | setup_clear_cpu_cap(X86_FEATURE_CLFLUSH); |
da4aaa7d | 1689 | setup_clear_cpu_cap(X86_FEATURE_CLFLUSHOPT); |
191679fd AK |
1690 | return 1; |
1691 | } | |
1692 | __setup("noclflush", setup_noclflush); | |
1693 | ||
148f9bb8 | 1694 | void print_cpu_info(struct cpuinfo_x86 *c) |
1da177e4 | 1695 | { |
02dde8b4 | 1696 | const char *vendor = NULL; |
1da177e4 | 1697 | |
0f3fa48a | 1698 | if (c->x86_vendor < X86_VENDOR_NUM) { |
1da177e4 | 1699 | vendor = this_cpu->c_vendor; |
0f3fa48a IM |
1700 | } else { |
1701 | if (c->cpuid_level >= 0) | |
1702 | vendor = c->x86_vendor_id; | |
1703 | } | |
1da177e4 | 1704 | |
bd32a8cf | 1705 | if (vendor && !strstr(c->x86_model_id, vendor)) |
1b74dde7 | 1706 | pr_cont("%s ", vendor); |
1da177e4 | 1707 | |
9d31d35b | 1708 | if (c->x86_model_id[0]) |
1b74dde7 | 1709 | pr_cont("%s", c->x86_model_id); |
1da177e4 | 1710 | else |
1b74dde7 | 1711 | pr_cont("%d86", c->x86); |
1da177e4 | 1712 | |
1b74dde7 | 1713 | pr_cont(" (family: 0x%x, model: 0x%x", c->x86, c->x86_model); |
924e101a | 1714 | |
b399151c JZ |
1715 | if (c->x86_stepping || c->cpuid_level >= 0) |
1716 | pr_cont(", stepping: 0x%x)\n", c->x86_stepping); | |
1da177e4 | 1717 | else |
1b74dde7 | 1718 | pr_cont(")\n"); |
1da177e4 LT |
1719 | } |
1720 | ||
0c2a3913 AK |
1721 | /* |
1722 | * clearcpuid= was already parsed in fpu__init_parse_early_param. | |
1723 | * But we need to keep a dummy __setup around otherwise it would | |
1724 | * show up as an environment variable for init. | |
1725 | */ | |
1726 | static __init int setup_clearcpuid(char *arg) | |
ac72e788 | 1727 | { |
ac72e788 AK |
1728 | return 1; |
1729 | } | |
0c2a3913 | 1730 | __setup("clearcpuid=", setup_clearcpuid); |
ac72e788 | 1731 | |
d5494d4f | 1732 | #ifdef CONFIG_X86_64 |
e6401c13 AL |
1733 | DEFINE_PER_CPU_FIRST(struct fixed_percpu_data, |
1734 | fixed_percpu_data) __aligned(PAGE_SIZE) __visible; | |
1735 | EXPORT_PER_CPU_SYMBOL_GPL(fixed_percpu_data); | |
0f3fa48a | 1736 | |
bdf977b3 | 1737 | /* |
a7fcf28d AL |
1738 | * The following percpu variables are hot. Align current_task to |
1739 | * cacheline size such that they fall in the same cacheline. | |
bdf977b3 TH |
1740 | */ |
1741 | DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned = | |
1742 | &init_task; | |
1743 | EXPORT_PER_CPU_SYMBOL(current_task); | |
d5494d4f | 1744 | |
951c2a51 | 1745 | DEFINE_PER_CPU(void *, hardirq_stack_ptr); |
e7f89001 | 1746 | DEFINE_PER_CPU(bool, hardirq_stack_inuse); |
d5494d4f | 1747 | |
c2daa3be PZ |
1748 | DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT; |
1749 | EXPORT_PER_CPU_SYMBOL(__preempt_count); | |
1750 | ||
1591584e LJ |
1751 | DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack) = TOP_OF_INIT_STACK; |
1752 | ||
d5494d4f YL |
1753 | /* May not be marked __init: used by software suspend */ |
1754 | void syscall_init(void) | |
1da177e4 | 1755 | { |
31ac34ca | 1756 | wrmsr(MSR_STAR, 0, (__USER32_CS << 16) | __KERNEL_CS); |
bf904d27 | 1757 | wrmsrl(MSR_LSTAR, (unsigned long)entry_SYSCALL_64); |
d56fe4bf IM |
1758 | |
1759 | #ifdef CONFIG_IA32_EMULATION | |
47edb651 | 1760 | wrmsrl(MSR_CSTAR, (unsigned long)entry_SYSCALL_compat); |
a76c7f46 | 1761 | /* |
487d1edb DV |
1762 | * This only works on Intel CPUs. |
1763 | * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP. | |
1764 | * This does not cause SYSENTER to jump to the wrong location, because | |
1765 | * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit). | |
a76c7f46 DV |
1766 | */ |
1767 | wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS); | |
8e6b65a1 | 1768 | wrmsrl_safe(MSR_IA32_SYSENTER_ESP, |
1769 | (unsigned long)(cpu_entry_stack(smp_processor_id()) + 1)); | |
4c8cd0c5 | 1770 | wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)entry_SYSENTER_compat); |
d56fe4bf | 1771 | #else |
47edb651 | 1772 | wrmsrl(MSR_CSTAR, (unsigned long)ignore_sysret); |
6b51311c | 1773 | wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG); |
d56fe4bf IM |
1774 | wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL); |
1775 | wrmsrl_safe(MSR_IA32_SYSENTER_EIP, 0ULL); | |
d5494d4f | 1776 | #endif |
03ae5768 | 1777 | |
d5494d4f YL |
1778 | /* Flags to clear on syscall */ |
1779 | wrmsrl(MSR_SYSCALL_MASK, | |
63bcff2a | 1780 | X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF| |
8c7aa698 | 1781 | X86_EFLAGS_IOPL|X86_EFLAGS_AC|X86_EFLAGS_NT); |
1da177e4 | 1782 | } |
62111195 | 1783 | |
0f3fa48a | 1784 | #else /* CONFIG_X86_64 */ |
d5494d4f | 1785 | |
bdf977b3 TH |
1786 | DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task; |
1787 | EXPORT_PER_CPU_SYMBOL(current_task); | |
c2daa3be PZ |
1788 | DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT; |
1789 | EXPORT_PER_CPU_SYMBOL(__preempt_count); | |
bdf977b3 | 1790 | |
a7fcf28d AL |
1791 | /* |
1792 | * On x86_32, vm86 modifies tss.sp0, so sp0 isn't a reliable way to find | |
1793 | * the top of the kernel stack. Use an extra percpu variable to track the | |
1794 | * top of the kernel stack directly. | |
1795 | */ | |
1796 | DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack) = | |
1797 | (unsigned long)&init_thread_union + THREAD_SIZE; | |
1798 | EXPORT_PER_CPU_SYMBOL(cpu_current_top_of_stack); | |
1799 | ||
050e9baa | 1800 | #ifdef CONFIG_STACKPROTECTOR |
53f82452 | 1801 | DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary); |
60a5317f | 1802 | #endif |
d5494d4f | 1803 | |
0f3fa48a | 1804 | #endif /* CONFIG_X86_64 */ |
c5413fbe | 1805 | |
9766cdbc JSR |
1806 | /* |
1807 | * Clear all 6 debug registers: | |
1808 | */ | |
1809 | static void clear_all_debug_regs(void) | |
1810 | { | |
1811 | int i; | |
1812 | ||
1813 | for (i = 0; i < 8; i++) { | |
1814 | /* Ignore db4, db5 */ | |
1815 | if ((i == 4) || (i == 5)) | |
1816 | continue; | |
1817 | ||
1818 | set_debugreg(0, i); | |
1819 | } | |
1820 | } | |
c5413fbe | 1821 | |
0bb9fef9 JW |
1822 | #ifdef CONFIG_KGDB |
1823 | /* | |
1824 | * Restore debug regs if using kgdbwait and you have a kernel debugger | |
1825 | * connection established. | |
1826 | */ | |
1827 | static void dbg_restore_debug_regs(void) | |
1828 | { | |
1829 | if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break)) | |
1830 | arch_kgdb_ops.correct_hw_break(); | |
1831 | } | |
1832 | #else /* ! CONFIG_KGDB */ | |
1833 | #define dbg_restore_debug_regs() | |
1834 | #endif /* ! CONFIG_KGDB */ | |
1835 | ||
ce4b1b16 IM |
1836 | static void wait_for_master_cpu(int cpu) |
1837 | { | |
1838 | #ifdef CONFIG_SMP | |
1839 | /* | |
1840 | * wait for ACK from master CPU before continuing | |
1841 | * with AP initialization | |
1842 | */ | |
1843 | WARN_ON(cpumask_test_and_set_cpu(cpu, cpu_initialized_mask)); | |
1844 | while (!cpumask_test_cpu(cpu, cpu_callout_mask)) | |
1845 | cpu_relax(); | |
1846 | #endif | |
1847 | } | |
1848 | ||
b2e2ba57 | 1849 | #ifdef CONFIG_X86_64 |
505b7899 | 1850 | static inline void setup_getcpu(int cpu) |
b2e2ba57 | 1851 | { |
22245bdf | 1852 | unsigned long cpudata = vdso_encode_cpunode(cpu, early_cpu_to_node(cpu)); |
b2e2ba57 CB |
1853 | struct desc_struct d = { }; |
1854 | ||
67e87d43 | 1855 | if (boot_cpu_has(X86_FEATURE_RDTSCP)) |
b2e2ba57 CB |
1856 | write_rdtscp_aux(cpudata); |
1857 | ||
1858 | /* Store CPU and node number in limit. */ | |
1859 | d.limit0 = cpudata; | |
1860 | d.limit1 = cpudata >> 16; | |
1861 | ||
1862 | d.type = 5; /* RO data, expand down, accessed */ | |
1863 | d.dpl = 3; /* Visible to user code */ | |
1864 | d.s = 1; /* Not a system segment */ | |
1865 | d.p = 1; /* Present */ | |
1866 | d.d = 1; /* 32-bit */ | |
1867 | ||
22245bdf | 1868 | write_gdt_entry(get_cpu_gdt_rw(cpu), GDT_ENTRY_CPUNODE, &d, DESCTYPE_S); |
b2e2ba57 | 1869 | } |
505b7899 TG |
1870 | |
1871 | static inline void ucode_cpu_init(int cpu) | |
1872 | { | |
1873 | if (cpu) | |
1874 | load_ucode_ap(); | |
1875 | } | |
1876 | ||
1877 | static inline void tss_setup_ist(struct tss_struct *tss) | |
1878 | { | |
1879 | /* Set up the per-CPU TSS IST stacks */ | |
1880 | tss->x86_tss.ist[IST_INDEX_DF] = __this_cpu_ist_top_va(DF); | |
1881 | tss->x86_tss.ist[IST_INDEX_NMI] = __this_cpu_ist_top_va(NMI); | |
1882 | tss->x86_tss.ist[IST_INDEX_DB] = __this_cpu_ist_top_va(DB); | |
1883 | tss->x86_tss.ist[IST_INDEX_MCE] = __this_cpu_ist_top_va(MCE); | |
02772fb9 JR |
1884 | /* Only mapped when SEV-ES is active */ |
1885 | tss->x86_tss.ist[IST_INDEX_VC] = __this_cpu_ist_top_va(VC); | |
505b7899 TG |
1886 | } |
1887 | ||
505b7899 TG |
1888 | #else /* CONFIG_X86_64 */ |
1889 | ||
1890 | static inline void setup_getcpu(int cpu) { } | |
1891 | ||
1892 | static inline void ucode_cpu_init(int cpu) | |
1893 | { | |
1894 | show_ucode_info_early(); | |
1895 | } | |
1896 | ||
1897 | static inline void tss_setup_ist(struct tss_struct *tss) { } | |
1898 | ||
505b7899 | 1899 | #endif /* !CONFIG_X86_64 */ |
b2e2ba57 | 1900 | |
111e7b15 TG |
1901 | static inline void tss_setup_io_bitmap(struct tss_struct *tss) |
1902 | { | |
1903 | tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET_INVALID; | |
1904 | ||
1905 | #ifdef CONFIG_X86_IOPL_IOPERM | |
1906 | tss->io_bitmap.prev_max = 0; | |
1907 | tss->io_bitmap.prev_sequence = 0; | |
1908 | memset(tss->io_bitmap.bitmap, 0xff, sizeof(tss->io_bitmap.bitmap)); | |
1909 | /* | |
1910 | * Invalidate the extra array entry past the end of the all | |
1911 | * permission bitmap as required by the hardware. | |
1912 | */ | |
1913 | tss->io_bitmap.mapall[IO_BITMAP_LONGS] = ~0UL; | |
b2e2ba57 | 1914 | #endif |
111e7b15 | 1915 | } |
b2e2ba57 | 1916 | |
520d0308 JR |
1917 | /* |
1918 | * Setup everything needed to handle exceptions from the IDT, including the IST | |
1919 | * exceptions which use paranoid_entry(). | |
1920 | */ | |
1921 | void cpu_init_exception_handling(void) | |
1922 | { | |
1923 | struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw); | |
1924 | int cpu = raw_smp_processor_id(); | |
1925 | ||
1926 | /* paranoid_entry() gets the CPU number from the GDT */ | |
1927 | setup_getcpu(cpu); | |
1928 | ||
1929 | /* IST vectors need TSS to be set up. */ | |
1930 | tss_setup_ist(tss); | |
1931 | tss_setup_io_bitmap(tss); | |
1932 | set_tss_desc(cpu, &get_cpu_entry_area(cpu)->tss.x86_tss); | |
1933 | ||
1934 | load_TR_desc(); | |
1935 | ||
1936 | /* Finally load the IDT */ | |
1937 | load_current_idt(); | |
1938 | } | |
1939 | ||
d2cbcc49 RR |
1940 | /* |
1941 | * cpu_init() initializes state that is per-CPU. Some data is already | |
1942 | * initialized (naturally) in the bootstrap process, such as the GDT | |
1943 | * and IDT. We reload them nevertheless, this function acts as a | |
1944 | * 'CPU state barrier', nothing should get across. | |
1945 | */ | |
148f9bb8 | 1946 | void cpu_init(void) |
1ba76586 | 1947 | { |
505b7899 TG |
1948 | struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw); |
1949 | struct task_struct *cur = current; | |
f6ef7322 | 1950 | int cpu = raw_smp_processor_id(); |
1ba76586 | 1951 | |
ce4b1b16 IM |
1952 | wait_for_master_cpu(cpu); |
1953 | ||
505b7899 | 1954 | ucode_cpu_init(cpu); |
0f3fa48a | 1955 | |
e7a22c1e | 1956 | #ifdef CONFIG_NUMA |
27fd185f | 1957 | if (this_cpu_read(numa_node) == 0 && |
e534c7c5 LS |
1958 | early_cpu_to_node(cpu) != NUMA_NO_NODE) |
1959 | set_numa_node(early_cpu_to_node(cpu)); | |
e7a22c1e | 1960 | #endif |
b2e2ba57 | 1961 | setup_getcpu(cpu); |
1ba76586 | 1962 | |
2eaad1fd | 1963 | pr_debug("Initializing CPU#%d\n", cpu); |
1ba76586 | 1964 | |
505b7899 TG |
1965 | if (IS_ENABLED(CONFIG_X86_64) || cpu_feature_enabled(X86_FEATURE_VME) || |
1966 | boot_cpu_has(X86_FEATURE_TSC) || boot_cpu_has(X86_FEATURE_DE)) | |
1967 | cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE); | |
1ba76586 YL |
1968 | |
1969 | /* | |
1970 | * Initialize the per-CPU GDT with the boot GDT, | |
1971 | * and set up the GDT descriptor: | |
1972 | */ | |
552be871 | 1973 | switch_to_new_gdt(cpu); |
cf910e83 | 1974 | load_current_idt(); |
1ba76586 | 1975 | |
505b7899 TG |
1976 | if (IS_ENABLED(CONFIG_X86_64)) { |
1977 | loadsegment(fs, 0); | |
1978 | memset(cur->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8); | |
1979 | syscall_init(); | |
1ba76586 | 1980 | |
505b7899 TG |
1981 | wrmsrl(MSR_FS_BASE, 0); |
1982 | wrmsrl(MSR_KERNEL_GS_BASE, 0); | |
1983 | barrier(); | |
1ba76586 | 1984 | |
505b7899 | 1985 | x2apic_setup(); |
1ba76586 YL |
1986 | } |
1987 | ||
f1f10076 | 1988 | mmgrab(&init_mm); |
505b7899 TG |
1989 | cur->active_mm = &init_mm; |
1990 | BUG_ON(cur->mm); | |
72c0098d | 1991 | initialize_tlbstate_and_flush(); |
505b7899 | 1992 | enter_lazy_tlb(&init_mm, cur); |
1ba76586 | 1993 | |
505b7899 TG |
1994 | /* Initialize the TSS. */ |
1995 | tss_setup_ist(tss); | |
111e7b15 | 1996 | tss_setup_io_bitmap(tss); |
72f5e08d | 1997 | set_tss_desc(cpu, &get_cpu_entry_area(cpu)->tss.x86_tss); |
505b7899 | 1998 | |
1ba76586 | 1999 | load_TR_desc(); |
505b7899 TG |
2000 | /* |
2001 | * sp0 points to the entry trampoline stack regardless of what task | |
2002 | * is running. | |
2003 | */ | |
4fe2d8b1 | 2004 | load_sp0((unsigned long)(cpu_entry_stack(cpu) + 1)); |
20bb8344 | 2005 | |
37868fe1 | 2006 | load_mm_ldt(&init_mm); |
1ba76586 | 2007 | |
0bb9fef9 JW |
2008 | clear_all_debug_regs(); |
2009 | dbg_restore_debug_regs(); | |
1ba76586 | 2010 | |
dc4e0021 | 2011 | doublefault_init_cpu_tss(); |
505b7899 | 2012 | |
21c4cd10 | 2013 | fpu__init_cpu(); |
1ba76586 | 2014 | |
1ba76586 YL |
2015 | if (is_uv_system()) |
2016 | uv_cpu_init(); | |
69218e47 | 2017 | |
69218e47 | 2018 | load_fixmap_gdt(cpu); |
1ba76586 YL |
2019 | } |
2020 | ||
1008c52c BP |
2021 | /* |
2022 | * The microcode loader calls this upon late microcode load to recheck features, | |
2023 | * only when microcode has been updated. Caller holds microcode_mutex and CPU | |
2024 | * hotplug lock. | |
2025 | */ | |
2026 | void microcode_check(void) | |
2027 | { | |
42ca8082 BP |
2028 | struct cpuinfo_x86 info; |
2029 | ||
1008c52c | 2030 | perf_check_microcode(); |
42ca8082 BP |
2031 | |
2032 | /* Reload CPUID max function as it might've changed. */ | |
2033 | info.cpuid_level = cpuid_eax(0); | |
2034 | ||
2035 | /* | |
2036 | * Copy all capability leafs to pick up the synthetic ones so that | |
2037 | * memcmp() below doesn't fail on that. The ones coming from CPUID will | |
2038 | * get overwritten in get_cpu_cap(). | |
2039 | */ | |
2040 | memcpy(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability)); | |
2041 | ||
2042 | get_cpu_cap(&info); | |
2043 | ||
2044 | if (!memcmp(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability))) | |
2045 | return; | |
2046 | ||
2047 | pr_warn("x86/CPU: CPU features have changed after loading microcode, but might not take effect.\n"); | |
2048 | pr_warn("x86/CPU: Please consider either early loading through initrd/built-in or a potential BIOS update.\n"); | |
1008c52c | 2049 | } |
9c92374b TG |
2050 | |
2051 | /* | |
2052 | * Invoked from core CPU hotplug code after hotplug operations | |
2053 | */ | |
2054 | void arch_smt_update(void) | |
2055 | { | |
2056 | /* Handle the speculative execution misfeatures */ | |
2057 | cpu_bugs_smt_update(); | |
6a1cb5f5 TG |
2058 | /* Check whether IPI broadcasting can be enabled */ |
2059 | apic_smt_update(); | |
9c92374b | 2060 | } |