1 | /* SPDX-License-Identifier: GPL-2.0 */ |
2 | /* |
3 | * |
4 | * Copyright (C) 1991, 1992 Linus Torvalds |
5 | * |
6 | * Enhanced CPU detection and feature setting code by Mike Jagdis |
7 | * and Martin Mares, November 1997. |
8 | */ |
9 | |
10 | .text |
11 | #include <linux/export.h> |
12 | #include <linux/threads.h> |
13 | #include <linux/init.h> |
14 | #include <linux/linkage.h> |
15 | #include <asm/segment.h> |
16 | #include <asm/page_types.h> |
17 | #include <asm/pgtable_types.h> |
18 | #include <asm/cache.h> |
19 | #include <asm/thread_info.h> |
20 | #include <asm/asm-offsets.h> |
21 | #include <asm/setup.h> |
22 | #include <asm/processor-flags.h> |
23 | #include <asm/msr-index.h> |
24 | #include <asm/cpufeatures.h> |
25 | #include <asm/percpu.h> |
26 | #include <asm/nops.h> |
27 | #include <asm/nospec-branch.h> |
28 | #include <asm/bootparam.h> |
29 | #include <asm/pgtable_32.h> |
30 | |
31 | /* Physical address */ |
32 | #define pa(X) ((X) - __PAGE_OFFSET) |
33 | |
34 | /* |
35 | * References to members of the new_cpu_data structure. |
36 | */ |
37 | |
38 | #define X86 new_cpu_data+CPUINFO_x86 |
39 | #define X86_VENDOR new_cpu_data+CPUINFO_x86_vendor |
40 | #define X86_MODEL new_cpu_data+CPUINFO_x86_model |
41 | #define X86_STEPPING new_cpu_data+CPUINFO_x86_stepping |
42 | #define X86_HARD_MATH new_cpu_data+CPUINFO_hard_math |
43 | #define X86_CPUID new_cpu_data+CPUINFO_cpuid_level |
44 | #define X86_CAPABILITY new_cpu_data+CPUINFO_x86_capability |
45 | #define X86_VENDOR_ID new_cpu_data+CPUINFO_x86_vendor_id |
46 | |
47 | |
48 | #define SIZEOF_PTREGS 17*4 |
49 | |
50 | /* |
51 | * Worst-case size of the kernel mapping we need to make: |
52 | * a relocatable kernel can live anywhere in lowmem, so we need to be able |
53 | * to map all of lowmem. |
54 | */ |
55 | KERNEL_PAGES = LOWMEM_PAGES |
56 | |
57 | INIT_MAP_SIZE = PAGE_TABLE_SIZE(KERNEL_PAGES) * PAGE_SIZE |
58 | RESERVE_BRK(pagetables, INIT_MAP_SIZE) |
59 | |
60 | /* |
61 | * 32-bit kernel entrypoint; only used by the boot CPU. On entry, |
62 | * %esi points to the real-mode code as a 32-bit pointer. |
63 | * CS and DS must be 4 GB flat segments, but we don't depend on |
64 | * any particular GDT layout, because we load our own as soon as we |
65 | * can. |
66 | */ |
67 | __HEAD |
68 | SYM_CODE_START(startup_32) |
69 | movl pa(initial_stack),%ecx |
70 | |
71 | /* |
72 | * Set segments to known values. |
73 | */ |
74 | lgdt pa(boot_gdt_descr) |
75 | movl $(__BOOT_DS),%eax |
76 | movl %eax,%ds |
77 | movl %eax,%es |
78 | movl %eax,%fs |
79 | movl %eax,%gs |
80 | movl %eax,%ss |
81 | leal -__PAGE_OFFSET(%ecx),%esp |
82 | |
83 | /* |
84 | * Clear BSS first so that there are no surprises... |
85 | */ |
86 | cld |
87 | xorl %eax,%eax |
88 | movl $pa(__bss_start),%edi |
89 | movl $pa(__bss_stop),%ecx |
90 | subl %edi,%ecx |
91 | shrl $2,%ecx |
92 | rep ; stosl |
93 | /* |
94 | * Copy bootup parameters out of the way. |
95 | * Note: %esi still has the pointer to the real-mode data. |
96 | * With the kexec as boot loader, parameter segment might be loaded beyond |
97 | * kernel image and might not even be addressable by early boot page tables. |
98 | * (kexec on panic case). Hence copy out the parameters before initializing |
99 | * page tables. |
100 | */ |
101 | movl $pa(boot_params),%edi |
102 | movl $(PARAM_SIZE/4),%ecx |
103 | cld |
104 | rep |
105 | movsl |
106 | movl pa(boot_params) + NEW_CL_POINTER,%esi |
107 | andl %esi,%esi |
108 | jz 1f # No command line |
109 | movl $pa(boot_command_line),%edi |
110 | movl $(COMMAND_LINE_SIZE/4),%ecx |
111 | rep |
112 | movsl |
113 | 1: |
114 | |
115 | #ifdef CONFIG_OLPC |
116 | /* save OFW's pgdir table for later use when calling into OFW */ |
117 | movl %cr3, %eax |
118 | movl %eax, pa(olpc_ofw_pgd) |
119 | #endif |
120 | |
121 | /* Create early pagetables. */ |
122 | call mk_early_pgtbl_32 |
123 | |
124 | /* Do early initialization of the fixmap area */ |
125 | movl $pa(initial_pg_fixmap)+PDE_IDENT_ATTR,%eax |
126 | #ifdef CONFIG_X86_PAE |
127 | #define KPMDS (((-__PAGE_OFFSET) >> 30) & 3) /* Number of kernel PMDs */ |
128 | movl %eax,pa(initial_pg_pmd+0x1000*KPMDS-8) |
129 | #else |
130 | movl %eax,pa(initial_page_table+0xffc) |
131 | #endif |
132 | |
133 | jmp .Ldefault_entry |
134 | SYM_CODE_END(startup_32) |
135 | |
136 | /* |
137 | * Non-boot CPU entry point; entered from trampoline.S |
138 | * We can't lgdt here, because lgdt itself uses a data segment, but |
139 | * we know the trampoline has already loaded the boot_gdt for us. |
140 | * |
141 | * If cpu hotplug is not supported then this code can go in init section |
142 | * which will be freed later |
143 | */ |
144 | SYM_FUNC_START(startup_32_smp) |
145 | cld |
146 | movl $(__BOOT_DS),%eax |
147 | movl %eax,%ds |
148 | movl %eax,%es |
149 | movl %eax,%fs |
150 | movl %eax,%gs |
151 | movl pa(initial_stack),%ecx |
152 | movl %eax,%ss |
153 | leal -__PAGE_OFFSET(%ecx),%esp |
154 | |
155 | .Ldefault_entry: |
156 | movl $(CR0_STATE & ~X86_CR0_PG),%eax |
157 | movl %eax,%cr0 |
158 | |
159 | /* |
160 | * We want to start out with EFLAGS unambiguously cleared. Some BIOSes leave |
161 | * bits like NT set. This would confuse the debugger if this code is traced. So |
162 | * initialize them properly now before switching to protected mode. That means |
163 | * DF in particular (even though we have cleared it earlier after copying the |
164 | * command line) because GCC expects it. |
165 | */ |
166 | pushl $0 |
167 | popfl |
168 | |
169 | /* |
170 | * New page tables may be in 4Mbyte page mode and may be using the global pages. |
171 | * |
172 | * NOTE! If we are on a 486 we may have no cr4 at all! Specifically, cr4 exists |
173 | * if and only if CPUID exists and has flags other than the FPU flag set. |
174 | */ |
175 | movl $-1,pa(X86_CPUID) # preset CPUID level |
176 | movl $X86_EFLAGS_ID,%ecx |
177 | pushl %ecx |
178 | popfl # set EFLAGS=ID |
179 | pushfl |
180 | popl %eax # get EFLAGS |
181 | testl $X86_EFLAGS_ID,%eax # did EFLAGS.ID remained set? |
182 | jz .Lenable_paging # hw disallowed setting of ID bit |
183 | # which means no CPUID and no CR4 |
184 | |
185 | xorl %eax,%eax |
186 | cpuid |
187 | movl %eax,pa(X86_CPUID) # save largest std CPUID function |
188 | |
189 | movl $1,%eax |
190 | cpuid |
191 | andl $~1,%edx # Ignore CPUID.FPU |
192 | jz .Lenable_paging # No flags or only CPUID.FPU = no CR4 |
193 | |
194 | movl pa(mmu_cr4_features),%eax |
195 | movl %eax,%cr4 |
196 | |
197 | testb $X86_CR4_PAE, %al # check if PAE is enabled |
198 | jz .Lenable_paging |
199 | |
200 | /* Check if extended functions are implemented */ |
201 | movl $0x80000000, %eax |
202 | cpuid |
203 | /* Value must be in the range 0x80000001 to 0x8000ffff */ |
204 | subl $0x80000001, %eax |
205 | cmpl $(0x8000ffff-0x80000001), %eax |
206 | ja .Lenable_paging |
207 | |
208 | /* Clear bogus XD_DISABLE bits */ |
209 | call verify_cpu |
210 | |
211 | mov $0x80000001, %eax |
212 | cpuid |
213 | /* Execute Disable bit supported? */ |
214 | btl $(X86_FEATURE_NX & 31), %edx |
215 | jnc .Lenable_paging |
216 | |
217 | /* Setup EFER (Extended Feature Enable Register) */ |
218 | movl $MSR_EFER, %ecx |
219 | rdmsr |
220 | |
221 | btsl $_EFER_NX, %eax |
222 | /* Make changes effective */ |
223 | wrmsr |
224 | |
225 | .Lenable_paging: |
226 | |
227 | /* |
228 | * Enable paging |
229 | */ |
230 | movl $pa(initial_page_table), %eax |
231 | movl %eax,%cr3 /* set the page table pointer.. */ |
232 | movl $CR0_STATE,%eax |
233 | movl %eax,%cr0 /* ..and set paging (PG) bit */ |
234 | ljmp $__BOOT_CS,$1f /* Clear prefetch and normalize %eip */ |
235 | 1: |
236 | /* Shift the stack pointer to a virtual address */ |
237 | addl $__PAGE_OFFSET, %esp |
238 | |
239 | /* |
240 | * Check if it is 486 |
241 | */ |
242 | movb $4,X86 # at least 486 |
243 | cmpl $-1,X86_CPUID |
244 | je .Lis486 |
245 | |
246 | /* get vendor info */ |
247 | xorl %eax,%eax # call CPUID with 0 -> return vendor ID |
248 | cpuid |
249 | movl %eax,X86_CPUID # save CPUID level |
250 | movl %ebx,X86_VENDOR_ID # lo 4 chars |
251 | movl %edx,X86_VENDOR_ID+4 # next 4 chars |
252 | movl %ecx,X86_VENDOR_ID+8 # last 4 chars |
253 | |
254 | orl %eax,%eax # do we have processor info as well? |
255 | je .Lis486 |
256 | |
257 | movl $1,%eax # Use the CPUID instruction to get CPU type |
258 | cpuid |
259 | movb %al,%cl # save reg for future use |
260 | andb $0x0f,%ah # mask processor family |
261 | movb %ah,X86 |
262 | andb $0xf0,%al # mask model |
263 | shrb $4,%al |
264 | movb %al,X86_MODEL |
265 | andb $0x0f,%cl # mask mask revision |
266 | movb %cl,X86_STEPPING |
267 | movl %edx,X86_CAPABILITY |
268 | |
269 | .Lis486: |
270 | movl $0x50022,%ecx # set AM, WP, NE and MP |
271 | movl %cr0,%eax |
272 | andl $0x80000011,%eax # Save PG,PE,ET |
273 | orl %ecx,%eax |
274 | movl %eax,%cr0 |
275 | |
276 | lgdt early_gdt_descr |
277 | ljmp $(__KERNEL_CS),$1f |
278 | 1: movl $(__KERNEL_DS),%eax # reload all the segment registers |
279 | movl %eax,%ss # after changing gdt. |
280 | |
281 | movl $(__USER_DS),%eax # DS/ES contains default USER segment |
282 | movl %eax,%ds |
283 | movl %eax,%es |
284 | |
285 | movl $(__KERNEL_PERCPU), %eax |
286 | movl %eax,%fs # set this cpu's percpu |
287 | |
288 | xorl %eax,%eax |
289 | movl %eax,%gs # clear possible garbage in %gs |
290 | |
291 | xorl %eax,%eax # Clear LDT |
292 | lldt %ax |
293 | |
294 | call *(initial_code) |
295 | 1: jmp 1b |
296 | SYM_FUNC_END(startup_32_smp) |
297 | |
298 | #include "verify_cpu.S" |
299 | |
300 | __INIT |
301 | SYM_FUNC_START(early_idt_handler_array) |
302 | # 36(%esp) %eflags |
303 | # 32(%esp) %cs |
304 | # 28(%esp) %eip |
305 | # 24(%rsp) error code |
306 | i = 0 |
307 | .rept NUM_EXCEPTION_VECTORS |
308 | .if ((EXCEPTION_ERRCODE_MASK >> i) & 1) == 0 |
309 | pushl $0 # Dummy error code, to make stack frame uniform |
310 | .endif |
311 | pushl $i # 20(%esp) Vector number |
312 | jmp early_idt_handler_common |
313 | i = i + 1 |
314 | .fill early_idt_handler_array + i*EARLY_IDT_HANDLER_SIZE - ., 1, 0xcc |
315 | .endr |
316 | SYM_FUNC_END(early_idt_handler_array) |
317 | |
318 | SYM_CODE_START_LOCAL(early_idt_handler_common) |
319 | /* |
320 | * The stack is the hardware frame, an error code or zero, and the |
321 | * vector number. |
322 | */ |
323 | cld |
324 | |
325 | incl %ss:early_recursion_flag |
326 | |
327 | /* The vector number is in pt_regs->gs */ |
328 | |
329 | cld |
330 | pushl %fs /* pt_regs->fs (__fsh varies by model) */ |
331 | pushl %es /* pt_regs->es (__esh varies by model) */ |
332 | pushl %ds /* pt_regs->ds (__dsh varies by model) */ |
333 | pushl %eax /* pt_regs->ax */ |
334 | pushl %ebp /* pt_regs->bp */ |
335 | pushl %edi /* pt_regs->di */ |
336 | pushl %esi /* pt_regs->si */ |
337 | pushl %edx /* pt_regs->dx */ |
338 | pushl %ecx /* pt_regs->cx */ |
339 | pushl %ebx /* pt_regs->bx */ |
340 | |
341 | /* Fix up DS and ES */ |
342 | movl $(__KERNEL_DS), %ecx |
343 | movl %ecx, %ds |
344 | movl %ecx, %es |
345 | |
346 | /* Load the vector number into EDX */ |
347 | movl PT_GS(%esp), %edx |
348 | |
349 | /* Load GS into pt_regs->gs (and maybe clobber __gsh) */ |
350 | movw %gs, PT_GS(%esp) |
351 | |
352 | movl %esp, %eax /* args are pt_regs (EAX), trapnr (EDX) */ |
353 | call early_fixup_exception |
354 | |
355 | popl %ebx /* pt_regs->bx */ |
356 | popl %ecx /* pt_regs->cx */ |
357 | popl %edx /* pt_regs->dx */ |
358 | popl %esi /* pt_regs->si */ |
359 | popl %edi /* pt_regs->di */ |
360 | popl %ebp /* pt_regs->bp */ |
361 | popl %eax /* pt_regs->ax */ |
362 | popl %ds /* pt_regs->ds (always ignores __dsh) */ |
363 | popl %es /* pt_regs->es (always ignores __esh) */ |
364 | popl %fs /* pt_regs->fs (always ignores __fsh) */ |
365 | popl %gs /* pt_regs->gs (always ignores __gsh) */ |
366 | decl %ss:early_recursion_flag |
367 | addl $4, %esp /* pop pt_regs->orig_ax */ |
368 | iret |
369 | SYM_CODE_END(early_idt_handler_common) |
370 | |
371 | /* This is the default interrupt "handler" :-) */ |
372 | SYM_FUNC_START(early_ignore_irq) |
373 | cld |
374 | #ifdef CONFIG_PRINTK |
375 | pushl %eax |
376 | pushl %ecx |
377 | pushl %edx |
378 | pushl %es |
379 | pushl %ds |
380 | movl $(__KERNEL_DS),%eax |
381 | movl %eax,%ds |
382 | movl %eax,%es |
383 | cmpl $2,early_recursion_flag |
384 | je hlt_loop |
385 | incl early_recursion_flag |
386 | pushl 16(%esp) |
387 | pushl 24(%esp) |
388 | pushl 32(%esp) |
389 | pushl 40(%esp) |
390 | pushl $int_msg |
391 | call _printk |
392 | |
393 | call dump_stack |
394 | |
395 | addl $(5*4),%esp |
396 | popl %ds |
397 | popl %es |
398 | popl %edx |
399 | popl %ecx |
400 | popl %eax |
401 | #endif |
402 | iret |
403 | |
404 | hlt_loop: |
405 | hlt |
406 | jmp hlt_loop |
407 | SYM_FUNC_END(early_ignore_irq) |
408 | |
409 | __INITDATA |
410 | .align 4 |
411 | SYM_DATA(early_recursion_flag, .long 0) |
412 | |
413 | __REFDATA |
414 | .align 4 |
415 | SYM_DATA(initial_code, .long i386_start_kernel) |
416 | |
417 | #ifdef CONFIG_PAGE_TABLE_ISOLATION |
418 | #define PGD_ALIGN (2 * PAGE_SIZE) |
419 | #define PTI_USER_PGD_FILL 1024 |
420 | #else |
421 | #define PGD_ALIGN (PAGE_SIZE) |
422 | #define PTI_USER_PGD_FILL 0 |
423 | #endif |
424 | /* |
425 | * BSS section |
426 | */ |
427 | __PAGE_ALIGNED_BSS |
428 | .align PGD_ALIGN |
429 | #ifdef CONFIG_X86_PAE |
430 | .globl initial_pg_pmd |
431 | initial_pg_pmd: |
432 | .fill 1024*KPMDS,4,0 |
433 | #else |
434 | .globl initial_page_table |
435 | initial_page_table: |
436 | .fill 1024,4,0 |
437 | #endif |
438 | .align PGD_ALIGN |
439 | initial_pg_fixmap: |
440 | .fill 1024,4,0 |
441 | .globl swapper_pg_dir |
442 | .align PGD_ALIGN |
443 | swapper_pg_dir: |
444 | .fill 1024,4,0 |
445 | .fill PTI_USER_PGD_FILL,4,0 |
446 | .globl empty_zero_page |
447 | empty_zero_page: |
448 | .fill 4096,1,0 |
449 | EXPORT_SYMBOL(empty_zero_page) |
450 | |
451 | /* |
452 | * This starts the data section. |
453 | */ |
454 | #ifdef CONFIG_X86_PAE |
455 | __PAGE_ALIGNED_DATA |
456 | /* Page-aligned for the benefit of paravirt? */ |
457 | .align PGD_ALIGN |
458 | SYM_DATA_START(initial_page_table) |
459 | .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0 /* low identity map */ |
460 | # if KPMDS == 3 |
461 | .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0 |
462 | .long pa(initial_pg_pmd+PGD_IDENT_ATTR+0x1000),0 |
463 | .long pa(initial_pg_pmd+PGD_IDENT_ATTR+0x2000),0 |
464 | # elif KPMDS == 2 |
465 | .long 0,0 |
466 | .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0 |
467 | .long pa(initial_pg_pmd+PGD_IDENT_ATTR+0x1000),0 |
468 | # elif KPMDS == 1 |
469 | .long 0,0 |
470 | .long 0,0 |
471 | .long pa(initial_pg_pmd+PGD_IDENT_ATTR),0 |
472 | # else |
473 | # error "Kernel PMDs should be 1, 2 or 3" |
474 | # endif |
475 | .align PAGE_SIZE /* needs to be page-sized too */ |
476 | |
477 | #ifdef CONFIG_PAGE_TABLE_ISOLATION |
478 | /* |
479 | * PTI needs another page so sync_initial_pagetable() works correctly |
480 | * and does not scribble over the data which is placed behind the |
481 | * actual initial_page_table. See clone_pgd_range(). |
482 | */ |
483 | .fill 1024, 4, 0 |
484 | #endif |
485 | |
486 | SYM_DATA_END(initial_page_table) |
487 | #endif |
488 | |
489 | .data |
490 | .balign 4 |
491 | /* |
492 | * The SIZEOF_PTREGS gap is a convention which helps the in-kernel unwinder |
493 | * reliably detect the end of the stack. |
494 | */ |
495 | SYM_DATA(initial_stack, |
496 | .long init_thread_union + THREAD_SIZE - |
497 | SIZEOF_PTREGS - TOP_OF_KERNEL_STACK_PADDING) |
498 | |
499 | __INITRODATA |
500 | int_msg: |
501 | .asciz "Unknown interrupt or fault at: %p %p %p\n" |
502 | |
503 | #include "../../x86/xen/xen-head.S" |
504 | |
505 | /* |
506 | * The IDT and GDT 'descriptors' are a strange 48-bit object |
507 | * only used by the lidt and lgdt instructions. They are not |
508 | * like usual segment descriptors - they consist of a 16-bit |
509 | * segment size, and 32-bit linear address value: |
510 | */ |
511 | |
512 | .data |
513 | ALIGN |
514 | # early boot GDT descriptor (must use 1:1 address mapping) |
515 | .word 0 # 32 bit align gdt_desc.address |
516 | SYM_DATA_START_LOCAL(boot_gdt_descr) |
517 | .word __BOOT_DS+7 |
518 | .long boot_gdt - __PAGE_OFFSET |
519 | SYM_DATA_END(boot_gdt_descr) |
520 | |
521 | # boot GDT descriptor (later on used by CPU#0): |
522 | .word 0 # 32 bit align gdt_desc.address |
523 | SYM_DATA_START(early_gdt_descr) |
524 | .word GDT_ENTRIES*8-1 |
525 | .long gdt_page /* Overwritten for secondary CPUs */ |
526 | SYM_DATA_END(early_gdt_descr) |
527 | |
528 | /* |
529 | * The boot_gdt must mirror the equivalent in setup.S and is |
530 | * used only for booting. |
531 | */ |
532 | .align L1_CACHE_BYTES |
533 | SYM_DATA_START(boot_gdt) |
534 | .fill GDT_ENTRY_BOOT_CS,8,0 |
535 | .quad 0x00cf9a000000ffff /* kernel 4GB code at 0x00000000 */ |
536 | .quad 0x00cf92000000ffff /* kernel 4GB data at 0x00000000 */ |
537 | SYM_DATA_END(boot_gdt) |
538 | |