1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * prepare to run common code
4 *
5 * Copyright (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
6 */
7
8 #define DISABLE_BRANCH_PROFILING
9 #include <linux/init.h>
10 #include <linux/linkage.h>
11 #include <linux/types.h>
12 #include <linux/kernel.h>
13 #include <linux/string.h>
14 #include <linux/percpu.h>
15 #include <linux/start_kernel.h>
16 #include <linux/io.h>
17 #include <linux/memblock.h>
18 #include <linux/mem_encrypt.h>
19
20 #include <asm/processor.h>
21 #include <asm/proto.h>
22 #include <asm/smp.h>
23 #include <asm/setup.h>
24 #include <asm/desc.h>
25 #include <asm/pgtable.h>
26 #include <asm/tlbflush.h>
27 #include <asm/sections.h>
28 #include <asm/kdebug.h>
29 #include <asm/e820/api.h>
30 #include <asm/bios_ebda.h>
31 #include <asm/bootparam_utils.h>
32 #include <asm/microcode.h>
33 #include <asm/kasan.h>
34 #include <asm/fixmap.h>
35
36 /*
37 * Manage page tables very early on.
38 */
39 extern pmd_t early_dynamic_pgts[EARLY_DYNAMIC_PAGE_TABLES][PTRS_PER_PMD];
40 static unsigned int __initdata next_early_pgt;
41 pmdval_t early_pmd_flags = __PAGE_KERNEL_LARGE & ~(_PAGE_GLOBAL | _PAGE_NX);
42
43 #define __head __section(.head.text)
44
fixup_pointer(void * ptr,unsigned long physaddr)45 static void __head *fixup_pointer(void *ptr, unsigned long physaddr)
46 {
47 return ptr - (void *)_text + (void *)physaddr;
48 }
49
__startup_64(unsigned long physaddr,struct boot_params * bp)50 unsigned long __head __startup_64(unsigned long physaddr,
51 struct boot_params *bp)
52 {
53 unsigned long load_delta, *p;
54 unsigned long pgtable_flags;
55 pgdval_t *pgd;
56 p4dval_t *p4d;
57 pudval_t *pud;
58 pmdval_t *pmd, pmd_entry;
59 int i;
60 unsigned int *next_pgt_ptr;
61
62 /* Is the address too large? */
63 if (physaddr >> MAX_PHYSMEM_BITS)
64 for (;;);
65
66 /*
67 * Compute the delta between the address I am compiled to run at
68 * and the address I am actually running at.
69 */
70 load_delta = physaddr - (unsigned long)(_text - __START_KERNEL_map);
71
72 /* Is the address not 2M aligned? */
73 if (load_delta & ~PMD_PAGE_MASK)
74 for (;;);
75
76 /* Activate Secure Memory Encryption (SME) if supported and enabled */
77 sme_enable(bp);
78
79 /* Include the SME encryption mask in the fixup value */
80 load_delta += sme_get_me_mask();
81
82 /* Fixup the physical addresses in the page table */
83
84 pgd = fixup_pointer(&early_top_pgt, physaddr);
85 pgd[pgd_index(__START_KERNEL_map)] += load_delta;
86
87 if (IS_ENABLED(CONFIG_X86_5LEVEL)) {
88 p4d = fixup_pointer(&level4_kernel_pgt, physaddr);
89 p4d[511] += load_delta;
90 }
91
92 pud = fixup_pointer(&level3_kernel_pgt, physaddr);
93 pud[510] += load_delta;
94 pud[511] += load_delta;
95
96 pmd = fixup_pointer(level2_fixmap_pgt, physaddr);
97 for (i = FIXMAP_PMD_TOP; i > FIXMAP_PMD_TOP - FIXMAP_PMD_NUM; i--)
98 pmd[i] += load_delta;
99
100 /*
101 * Set up the identity mapping for the switchover. These
102 * entries should *NOT* have the global bit set! This also
103 * creates a bunch of nonsense entries but that is fine --
104 * it avoids problems around wraparound.
105 */
106
107 next_pgt_ptr = fixup_pointer(&next_early_pgt, physaddr);
108 pud = fixup_pointer(early_dynamic_pgts[(*next_pgt_ptr)++], physaddr);
109 pmd = fixup_pointer(early_dynamic_pgts[(*next_pgt_ptr)++], physaddr);
110
111 pgtable_flags = _KERNPG_TABLE_NOENC + sme_get_me_mask();
112
113 if (IS_ENABLED(CONFIG_X86_5LEVEL)) {
114 p4d = fixup_pointer(early_dynamic_pgts[next_early_pgt++], physaddr);
115
116 i = (physaddr >> PGDIR_SHIFT) % PTRS_PER_PGD;
117 pgd[i + 0] = (pgdval_t)p4d + pgtable_flags;
118 pgd[i + 1] = (pgdval_t)p4d + pgtable_flags;
119
120 i = physaddr >> P4D_SHIFT;
121 p4d[(i + 0) % PTRS_PER_P4D] = (pgdval_t)pud + pgtable_flags;
122 p4d[(i + 1) % PTRS_PER_P4D] = (pgdval_t)pud + pgtable_flags;
123 } else {
124 i = (physaddr >> PGDIR_SHIFT) % PTRS_PER_PGD;
125 pgd[i + 0] = (pgdval_t)pud + pgtable_flags;
126 pgd[i + 1] = (pgdval_t)pud + pgtable_flags;
127 }
128
129 i = physaddr >> PUD_SHIFT;
130 pud[(i + 0) % PTRS_PER_PUD] = (pudval_t)pmd + pgtable_flags;
131 pud[(i + 1) % PTRS_PER_PUD] = (pudval_t)pmd + pgtable_flags;
132
133 pmd_entry = __PAGE_KERNEL_LARGE_EXEC & ~_PAGE_GLOBAL;
134 pmd_entry += sme_get_me_mask();
135 pmd_entry += physaddr;
136
137 for (i = 0; i < DIV_ROUND_UP(_end - _text, PMD_SIZE); i++) {
138 int idx = i + (physaddr >> PMD_SHIFT);
139
140 pmd[idx % PTRS_PER_PMD] = pmd_entry + i * PMD_SIZE;
141 }
142
143 /*
144 * Fixup the kernel text+data virtual addresses. Note that
145 * we might write invalid pmds, when the kernel is relocated
146 * cleanup_highmap() fixes this up along with the mappings
147 * beyond _end.
148 *
149 * Only the region occupied by the kernel image has so far
150 * been checked against the table of usable memory regions
151 * provided by the firmware, so invalidate pages outside that
152 * region. A page table entry that maps to a reserved area of
153 * memory would allow processor speculation into that area,
154 * and on some hardware (particularly the UV platform) even
155 * speculative access to some reserved areas is caught as an
156 * error, causing the BIOS to halt the system.
157 */
158
159 pmd = fixup_pointer(level2_kernel_pgt, physaddr);
160
161 /* invalidate pages before the kernel image */
162 for (i = 0; i < pmd_index((unsigned long)_text); i++)
163 pmd[i] &= ~_PAGE_PRESENT;
164
165 /* fixup pages that are part of the kernel image */
166 for (; i <= pmd_index((unsigned long)_end); i++)
167 if (pmd[i] & _PAGE_PRESENT)
168 pmd[i] += load_delta;
169
170 /* invalidate pages after the kernel image */
171 for (; i < PTRS_PER_PMD; i++)
172 pmd[i] &= ~_PAGE_PRESENT;
173
174 /*
175 * Fixup phys_base - remove the memory encryption mask to obtain
176 * the true physical address.
177 */
178 p = fixup_pointer(&phys_base, physaddr);
179 *p += load_delta - sme_get_me_mask();
180
181 /* Encrypt the kernel and related (if SME is active) */
182 sme_encrypt_kernel(bp);
183
184 /*
185 * Return the SME encryption mask (if SME is active) to be used as a
186 * modifier for the initial pgdir entry programmed into CR3.
187 */
188 return sme_get_me_mask();
189 }
190
__startup_secondary_64(void)191 unsigned long __startup_secondary_64(void)
192 {
193 /*
194 * Return the SME encryption mask (if SME is active) to be used as a
195 * modifier for the initial pgdir entry programmed into CR3.
196 */
197 return sme_get_me_mask();
198 }
199
200 /* Wipe all early page tables except for the kernel symbol map */
reset_early_page_tables(void)201 static void __init reset_early_page_tables(void)
202 {
203 memset(early_top_pgt, 0, sizeof(pgd_t)*(PTRS_PER_PGD-1));
204 next_early_pgt = 0;
205 write_cr3(__sme_pa_nodebug(early_top_pgt));
206 }
207
208 /* Create a new PMD entry */
__early_make_pgtable(unsigned long address,pmdval_t pmd)209 int __init __early_make_pgtable(unsigned long address, pmdval_t pmd)
210 {
211 unsigned long physaddr = address - __PAGE_OFFSET;
212 pgdval_t pgd, *pgd_p;
213 p4dval_t p4d, *p4d_p;
214 pudval_t pud, *pud_p;
215 pmdval_t *pmd_p;
216
217 /* Invalid address or early pgt is done ? */
218 if (physaddr >= MAXMEM || read_cr3_pa() != __pa_nodebug(early_top_pgt))
219 return -1;
220
221 again:
222 pgd_p = &early_top_pgt[pgd_index(address)].pgd;
223 pgd = *pgd_p;
224
225 /*
226 * The use of __START_KERNEL_map rather than __PAGE_OFFSET here is
227 * critical -- __PAGE_OFFSET would point us back into the dynamic
228 * range and we might end up looping forever...
229 */
230 if (!IS_ENABLED(CONFIG_X86_5LEVEL))
231 p4d_p = pgd_p;
232 else if (pgd)
233 p4d_p = (p4dval_t *)((pgd & PTE_PFN_MASK) + __START_KERNEL_map - phys_base);
234 else {
235 if (next_early_pgt >= EARLY_DYNAMIC_PAGE_TABLES) {
236 reset_early_page_tables();
237 goto again;
238 }
239
240 p4d_p = (p4dval_t *)early_dynamic_pgts[next_early_pgt++];
241 memset(p4d_p, 0, sizeof(*p4d_p) * PTRS_PER_P4D);
242 *pgd_p = (pgdval_t)p4d_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
243 }
244 p4d_p += p4d_index(address);
245 p4d = *p4d_p;
246
247 if (p4d)
248 pud_p = (pudval_t *)((p4d & PTE_PFN_MASK) + __START_KERNEL_map - phys_base);
249 else {
250 if (next_early_pgt >= EARLY_DYNAMIC_PAGE_TABLES) {
251 reset_early_page_tables();
252 goto again;
253 }
254
255 pud_p = (pudval_t *)early_dynamic_pgts[next_early_pgt++];
256 memset(pud_p, 0, sizeof(*pud_p) * PTRS_PER_PUD);
257 *p4d_p = (p4dval_t)pud_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
258 }
259 pud_p += pud_index(address);
260 pud = *pud_p;
261
262 if (pud)
263 pmd_p = (pmdval_t *)((pud & PTE_PFN_MASK) + __START_KERNEL_map - phys_base);
264 else {
265 if (next_early_pgt >= EARLY_DYNAMIC_PAGE_TABLES) {
266 reset_early_page_tables();
267 goto again;
268 }
269
270 pmd_p = (pmdval_t *)early_dynamic_pgts[next_early_pgt++];
271 memset(pmd_p, 0, sizeof(*pmd_p) * PTRS_PER_PMD);
272 *pud_p = (pudval_t)pmd_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
273 }
274 pmd_p[pmd_index(address)] = pmd;
275
276 return 0;
277 }
278
early_make_pgtable(unsigned long address)279 int __init early_make_pgtable(unsigned long address)
280 {
281 unsigned long physaddr = address - __PAGE_OFFSET;
282 pmdval_t pmd;
283
284 pmd = (physaddr & PMD_MASK) + early_pmd_flags;
285
286 return __early_make_pgtable(address, pmd);
287 }
288
289 /* Don't add a printk in there. printk relies on the PDA which is not initialized
290 yet. */
clear_bss(void)291 static void __init clear_bss(void)
292 {
293 memset(__bss_start, 0,
294 (unsigned long) __bss_stop - (unsigned long) __bss_start);
295 }
296
get_cmd_line_ptr(void)297 static unsigned long get_cmd_line_ptr(void)
298 {
299 unsigned long cmd_line_ptr = boot_params.hdr.cmd_line_ptr;
300
301 cmd_line_ptr |= (u64)boot_params.ext_cmd_line_ptr << 32;
302
303 return cmd_line_ptr;
304 }
305
copy_bootdata(char * real_mode_data)306 static void __init copy_bootdata(char *real_mode_data)
307 {
308 char * command_line;
309 unsigned long cmd_line_ptr;
310
311 /*
312 * If SME is active, this will create decrypted mappings of the
313 * boot data in advance of the copy operations.
314 */
315 sme_map_bootdata(real_mode_data);
316
317 memcpy(&boot_params, real_mode_data, sizeof boot_params);
318 sanitize_boot_params(&boot_params);
319 cmd_line_ptr = get_cmd_line_ptr();
320 if (cmd_line_ptr) {
321 command_line = __va(cmd_line_ptr);
322 memcpy(boot_command_line, command_line, COMMAND_LINE_SIZE);
323 }
324
325 /*
326 * The old boot data is no longer needed and won't be reserved,
327 * freeing up that memory for use by the system. If SME is active,
328 * we need to remove the mappings that were created so that the
329 * memory doesn't remain mapped as decrypted.
330 */
331 sme_unmap_bootdata(real_mode_data);
332 }
333
x86_64_start_kernel(char * real_mode_data)334 asmlinkage __visible void __init x86_64_start_kernel(char * real_mode_data)
335 {
336 /*
337 * Build-time sanity checks on the kernel image and module
338 * area mappings. (these are purely build-time and produce no code)
339 */
340 BUILD_BUG_ON(MODULES_VADDR < __START_KERNEL_map);
341 BUILD_BUG_ON(MODULES_VADDR - __START_KERNEL_map < KERNEL_IMAGE_SIZE);
342 BUILD_BUG_ON(MODULES_LEN + KERNEL_IMAGE_SIZE > 2*PUD_SIZE);
343 BUILD_BUG_ON((__START_KERNEL_map & ~PMD_MASK) != 0);
344 BUILD_BUG_ON((MODULES_VADDR & ~PMD_MASK) != 0);
345 BUILD_BUG_ON(!(MODULES_VADDR > __START_KERNEL));
346 BUILD_BUG_ON(!(((MODULES_END - 1) & PGDIR_MASK) ==
347 (__START_KERNEL & PGDIR_MASK)));
348 BUILD_BUG_ON(__fix_to_virt(__end_of_fixed_addresses) <= MODULES_END);
349
350 cr4_init_shadow();
351
352 /* Kill off the identity-map trampoline */
353 reset_early_page_tables();
354
355 clear_bss();
356
357 clear_page(init_top_pgt);
358
359 /*
360 * SME support may update early_pmd_flags to include the memory
361 * encryption mask, so it needs to be called before anything
362 * that may generate a page fault.
363 */
364 sme_early_init();
365
366 kasan_early_init();
367
368 idt_setup_early_handler();
369
370 copy_bootdata(__va(real_mode_data));
371
372 /*
373 * Load microcode early on BSP.
374 */
375 load_ucode_bsp();
376
377 /* set init_top_pgt kernel high mapping*/
378 init_top_pgt[511] = early_top_pgt[511];
379
380 x86_64_start_reservations(real_mode_data);
381 }
382
x86_64_start_reservations(char * real_mode_data)383 void __init x86_64_start_reservations(char *real_mode_data)
384 {
385 /* version is always not zero if it is copied */
386 if (!boot_params.hdr.version)
387 copy_bootdata(__va(real_mode_data));
388
389 x86_early_init_platform_quirks();
390
391 switch (boot_params.hdr.hardware_subarch) {
392 case X86_SUBARCH_INTEL_MID:
393 x86_intel_mid_early_setup();
394 break;
395 default:
396 break;
397 }
398
399 start_kernel();
400 }
401